{ "providers": { "openai": { "display_name": "OpenAI", "base_url": "https://api.openai.com/v1", "console_url": "https://platform.openai.com/api-keys", "no_china_endpoint": true, "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "gpt-6.2-sol", "gpt-6.2-astra", "gpt-6.1-terra", "gpt-6-luna", "gpt-5.6-sol" ], "notes": "Flagship: gpt-6.2-sol (1.05M ctx); most capable: gpt-6.2-astra; balanced: gpt-6.1-terra; fast/cheap: gpt-6-luna. GPT-5.4 and later (all GPT-6) are served over the Responses API: Chat Completions refuses function tools with reasoning on from GPT-5.4, and GPT-6 Astra / 6.1 Sol / 6.2 Sol also refuse reasoning effort 'none'. On Responses Orbital keeps reasoning on (store:false, encrypted reasoning items replayed with their tool calls, summaries shown as thinking). Older models (gpt-5.2 and earlier, gpt-4.x, o-series) stay on Chat Completions, where the reasoning trace is hidden. An uncatalogued gpt-6.x id resolves to the gpt-6 prefix entry and so also uses Responses. No separate China endpoint. https://developers.openai.com/api/docs/guides/migrate-to-responses", "pricing": { "gpt-5.2": { "input_per_1m": 1.75, "output_per_1m": 14.0 }, "gpt-5.1": { "input_per_1m": 1.25, "output_per_1m": 10.0 }, "gpt-5": { "input_per_1m": 1.25, "output_per_1m": 10.0 }, "gpt-5-mini": { "input_per_1m": 0.25, "output_per_1m": 2.0 }, "gpt-5-nano": { "input_per_1m": 0.05, "output_per_1m": 0.4 }, "gpt-4.1": { "input_per_1m": 2.0, "output_per_1m": 8.0 }, "gpt-4o": { "input_per_1m": 2.5, "output_per_1m": 10.0 }, "gpt-4o-mini": { "input_per_1m": 0.15, "output_per_1m": 0.6 }, "o3": { "input_per_1m": 2.0, "output_per_1m": 8.0 }, "o3-mini": { "input_per_1m": 1.1, "output_per_1m": 4.4 }, "o4-mini": { "input_per_1m": 1.1, "output_per_1m": 4.4 }, "o1": { "input_per_1m": 15.0, "output_per_1m": 60.0 }, "_default": { "input_per_1m": 2.5, "output_per_1m": 15.0 }, "gpt-5.5": { "input_per_1m": 5.0, "output_per_1m": 30.0 }, "gpt-5.5-pro": { "input_per_1m": 30.0, "output_per_1m": 180.0 }, "gpt-5.4": { "input_per_1m": 2.5, "output_per_1m": 15.0 }, "gpt-5.4-pro": { "input_per_1m": 30.0, "output_per_1m": 180.0 }, "gpt-5.4-mini": { "input_per_1m": 0.75, "output_per_1m": 4.5 }, "gpt-5.4-nano": { "input_per_1m": 0.2, "output_per_1m": 1.25 }, "gpt-5.3-codex": { "input_per_1m": 1.75, "output_per_1m": 14.0 }, "gpt-5.6-sol": { "input_per_1m": 5.0, "output_per_1m": 30.0 }, "gpt-5.6-terra": { "input_per_1m": 2.5, "output_per_1m": 15.0 }, "gpt-5.6-luna": { "input_per_1m": 1.0, "output_per_1m": 6.0 }, "gpt-5.6": { "input_per_1m": 5.0, "output_per_1m": 30.0 }, "gpt-6.2-sol": { "input_per_1m": 2.0, "output_per_1m": 10.0, "cached_input_per_1m": 0.1 }, "gpt-6.2-astra": { "input_per_1m": 8.0, "output_per_1m": 40.0, "cached_input_per_1m": 0.4 }, "gpt-6.1-sol": { "input_per_1m": 2.0, "output_per_1m": 10.0, "cached_input_per_1m": 0.1 }, "gpt-6.1-terra": { "input_per_1m": 1.0, "output_per_1m": 5.0, "cached_input_per_1m": 0.05 }, "gpt-6-astra": { "input_per_1m": 10.0, "output_per_1m": 50.0, "cached_input_per_1m": 1.0 }, "gpt-6-sol": { "input_per_1m": 2.0, "output_per_1m": 10.0, "cached_input_per_1m": 0.2 }, "gpt-6-luna": { "input_per_1m": 0.1, "output_per_1m": 0.5, "cached_input_per_1m": 0.01 }, "gpt-6*": { "input_per_1m": 2.0, "output_per_1m": 10.0, "cached_input_per_1m": 0.1 } }, "models": { "gpt-5.2": { "display_name": "GPT-5.2", "tier": "balanced", "context_window": 272000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "param:reasoning_effort=medium", "notes": "Reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "gpt-5.1": { "display_name": "GPT-5.1", "tier": "balanced", "context_window": 272000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "param:reasoning_effort=medium", "notes": "Reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "gpt-5": { "display_name": "GPT-5", "tier": "balanced", "context_window": 272000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "param:reasoning_effort=medium", "notes": "Reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "gpt-5-mini": { "display_name": "GPT-5 Mini", "tier": "fast", "context_window": 272000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "param:reasoning_effort=medium", "notes": "Reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "gpt-5-nano": { "display_name": "GPT-5 Nano", "tier": "fast", "context_window": 272000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "param:reasoning_effort=medium", "notes": "Reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "gpt-4.1": { "display_name": "GPT-4.1", "tier": "balanced", "context_window": 1047576, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model." } }, "gpt-4o": { "display_name": "GPT-4o", "tier": "balanced", "context_window": 128000, "max_output": 16384, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model." } }, "gpt-4o-mini": { "display_name": "GPT-4o Mini", "tier": "fast", "context_window": 128000, "max_output": 16384, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model." } }, "o3": { "display_name": "o3", "tier": "reasoning", "context_window": 200000, "max_output": 100000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "param:reasoning_effort=medium", "notes": "o3 reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "o3-mini": { "display_name": "o3 Mini", "tier": "reasoning", "context_window": 200000, "max_output": 100000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "param:reasoning_effort=medium", "notes": "o3-mini reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "o4-mini": { "display_name": "o4 Mini", "tier": "reasoning", "context_window": 200000, "max_output": 100000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "param:reasoning_effort=medium", "notes": "o4-mini reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "o1": { "display_name": "o1", "tier": "reasoning", "context_window": 200000, "max_output": 100000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "o1 reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "_default": { "context_window": 400000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true } }, "gpt-5.5": { "display_name": "GPT-5.5", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-5.5-pro": { "display_name": "GPT-5.5 Pro", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": false }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "medium", "high", "xhigh" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-5.4": { "display_name": "GPT-5.4", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-5.4-pro": { "display_name": "GPT-5.4 Pro", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "medium", "high", "xhigh" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-5.4-mini": { "display_name": "GPT-5.4 Mini", "tier": "fast", "context_window": 272000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-5.4-nano": { "display_name": "GPT-5.4 Nano", "tier": "fast", "context_window": 272000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-5.3-codex": { "display_name": "GPT-5.3 Codex", "tier": "balanced", "context_window": 272000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "param:reasoning_effort=medium", "notes": "Codex variant; reasoning hidden on Chat Completions. https://developers.openai.com/api/docs/guides/reasoning" } }, "gpt-5.6-sol": { "display_name": "GPT-5.6 Sol", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-5.6-terra": { "display_name": "GPT-5.6 Terra", "tier": "balanced", "context_window": 922000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-5.6-luna": { "display_name": "GPT-5.6 Luna", "tier": "fast", "context_window": 922000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-5.6": { "display_name": "GPT-5.6 (alias of Sol)", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses Alias id gpt-5.6 routes to gpt-5.6-sol." }, "sdk": "openai-responses" }, "gpt-6.2-sol": { "display_name": "GPT-6.2 Sol", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses Rejects reasoning effort 'none', so reasoning cannot be turned off. https://developers.openai.com/api/docs/models/gpt-6.2-sol" }, "sdk": "openai-responses" }, "gpt-6.2-astra": { "display_name": "GPT-6.2 Astra", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses Rejects reasoning effort 'none', so reasoning cannot be turned off." }, "sdk": "openai-responses" }, "gpt-6.1-sol": { "display_name": "GPT-6.1 Sol", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses Rejects reasoning effort 'none', so reasoning cannot be turned off." }, "sdk": "openai-responses" }, "gpt-6.1-terra": { "display_name": "GPT-6.1 Terra", "tier": "balanced", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-6-astra": { "display_name": "GPT-6 Astra", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses Rejects reasoning effort 'none', so reasoning cannot be turned off." }, "sdk": "openai-responses" }, "gpt-6-sol": { "display_name": "GPT-6 Sol", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-6-luna": { "display_name": "GPT-6 Luna", "tier": "fast", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses" }, "sdk": "openai-responses" }, "gpt-6*": { "display_name": "GPT-6 (uncatalogued)", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh", "max" ], "notes": "Served over the Responses API: from GPT-5.4 on, Chat Completions refuses function tools with reasoning on. Reasoning stays on; encrypted reasoning items are replayed (store:false) and summaries surface as thinking. https://developers.openai.com/api/docs/guides/migrate-to-responses Rejects reasoning effort 'none', so reasoning cannot be turned off. Family pattern (key ends in *, matches across versions): an uncatalogued gpt-6.x id resolves here, so it keeps the Responses endpoint instead of falling back to Chat Completions (flagship inheritance carries specs, never the endpoint). Effort list omits 'none' so the off switch is never sent to a model that may refuse it." }, "sdk": "openai-responses" } }, "cache": { "supported": "implicit", "control": "automatic", "default_ttl_min": 10, "extended_ttl_min": 1440, "extended_ttl_param": "model_only:gpt-5.5+", "isolation": "per_org", "isolation_confidence": "explicit", "min_tokens": 1024, "max_breakpoints": null, "observable": true, "observability_field": "usage.prompt_tokens_details.cached_tokens", "pollution_risk": "likely_safe", "notes": "Auto-cache >=1024 tokens. Use prompt_cache_key for routing stickiness; ~15 RPM/engine ceiling. Responses API caches reasoning persists, Chat Completions does not — meaningful caching with o-series effectively requires Responses API. https://developers.openai.com/api/docs/guides/prompt-caching" } }, "anthropic": { "display_name": "Anthropic", "base_url": "https://api.anthropic.com", "console_url": "https://console.anthropic.com/settings/keys", "no_china_endpoint": true, "currency": "USD", "supports_model_list": true, "model_list_endpoint": "/v1/models", "sdk": "anthropic", "suggested_models": [ "claude-fable-5", "claude-opus-4-8", "claude-sonnet-5", "claude-haiku-4-5" ], "notes": "Flagship: claude-opus-4-7 (1M ctx, adaptive thinking only). Balanced: claude-sonnet-4-6 (1M ctx, adaptive or legacy budget_tokens). Fast/cheap: claude-haiku-4-5 (200k ctx, manual budget_tokens). All three Claude 4.x models support extended thinking and EMIT `thinking` content blocks with opaque signatures. Echo-back of thinking blocks (signatures unmodified) is REQUIRED across tool-use turns; pure text continuations are lenient. Anthropic uses its own SDK, not /chat/completions. No separate China endpoint.", "pricing": { "claude-opus-4-6": { "input_per_1m": 5.0, "cached_input_per_1m": 0.5, "cache_write_per_1m": 6.25, "output_per_1m": 25.0 }, "claude-sonnet-4-6": { "input_per_1m": 3.0, "cached_input_per_1m": 0.3, "cache_write_per_1m": 3.75, "output_per_1m": 15.0 }, "claude-sonnet-4-5": { "input_per_1m": 3.0, "cached_input_per_1m": 0.3, "cache_write_per_1m": 3.75, "output_per_1m": 15.0 }, "claude-haiku-4-5": { "input_per_1m": 1.0, "cached_input_per_1m": 0.1, "cache_write_per_1m": 1.25, "output_per_1m": 5.0 }, "claude-opus-4-5": { "input_per_1m": 5.0, "cached_input_per_1m": 0.5, "cache_write_per_1m": 6.25, "output_per_1m": 25.0 }, "claude-opus-4-1": { "input_per_1m": 15.0, "cached_input_per_1m": 1.5, "cache_write_per_1m": 18.75, "output_per_1m": 75.0 }, "claude-sonnet-4": { "input_per_1m": 3.0, "cached_input_per_1m": 0.3, "cache_write_per_1m": 3.75, "output_per_1m": 15.0 }, "claude-opus-4": { "input_per_1m": 15.0, "cached_input_per_1m": 1.5, "cache_write_per_1m": 18.75, "output_per_1m": 75.0 }, "claude-3-haiku": { "input_per_1m": 0.25, "cached_input_per_1m": 0.025, "cache_write_per_1m": 0.3125, "output_per_1m": 1.25 }, "claude-3-5-sonnet": { "input_per_1m": 3.0, "cached_input_per_1m": 0.3, "cache_write_per_1m": 3.75, "output_per_1m": 15.0 }, "claude-3-opus": { "input_per_1m": 15.0, "cached_input_per_1m": 1.5, "cache_write_per_1m": 18.75, "output_per_1m": 75.0 }, "_default": { "input_per_1m": 3.0, "cached_input_per_1m": 0.3, "cache_write_per_1m": 3.75, "output_per_1m": 15.0 }, "claude-fable-5": { "input_per_1m": 10.0, "output_per_1m": 50.0 }, "claude-fable-5-1": { "input_per_1m": 10.0, "cached_input_per_1m": 0.25, "cache_write_per_1m": 12.5, "output_per_1m": 50.0 }, "claude-fable-5-2": { "input_per_1m": 8.0, "cached_input_per_1m": 0.2, "cache_write_per_1m": 10.0, "output_per_1m": 40.0 }, "claude-opus-4-8": { "input_per_1m": 5.0, "output_per_1m": 25.0 }, "claude-opus-4-7": { "input_per_1m": 5.0, "output_per_1m": 25.0 }, "claude-sonnet-5": { "input_per_1m": 3.0, "output_per_1m": 15.0 }, "claude-sonnet-5-5": { "input_per_1m": 2.0, "cached_input_per_1m": 0.1, "cache_write_per_1m": 2.5, "output_per_1m": 10.0 } }, "models": { "claude-opus-4-6": { "display_name": "Claude Opus 4.6", "tier": "flagship-legacy", "context_window": 1000000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'adaptive'}", "notes": "Legacy 4.6 opus with extended thinking." } }, "claude-sonnet-4-6": { "display_name": "Claude Sonnet 4.6", "tier": "balanced", "context_window": 1000000, "max_output": 64000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'adaptive'}", "notes": "Adaptive (recommended) or manual budget_tokens (deprecated). Echo-back of thinking blocks required during tool use. https://platform.claude.com/docs/en/build-with-claude/extended-thinking" } }, "claude-opus-4-5": { "display_name": "Claude Opus 4.5", "tier": "flagship", "context_window": 200000, "max_output": 16384, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Legacy 4.5 opus extended thinking with budget_tokens." } }, "claude-sonnet-4-5": { "display_name": "Claude Sonnet 4.5", "tier": "balanced", "context_window": 200000, "max_output": 16384, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'adaptive'}", "notes": "Legacy 4.5 sonnet with extended thinking; same echo-back rule." } }, "claude-haiku-4-5": { "display_name": "Claude Haiku 4.5", "tier": "fast", "context_window": 200000, "max_output": 32000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Manual extended thinking with budget_tokens. Echo-back required during tool use. https://platform.claude.com/docs/en/build-with-claude/extended-thinking" } }, "claude-sonnet-4": { "display_name": "Claude Sonnet 4", "tier": "balanced", "context_window": 200000, "max_output": 16384, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Legacy claude-sonnet-4 (May 2025); extended thinking with budget_tokens." } }, "claude-opus-4": { "display_name": "Claude Opus 4", "tier": "flagship", "context_window": 200000, "max_output": 16384, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Legacy claude-opus-4 (May 2025); extended thinking with budget_tokens." } }, "_default": { "context_window": 200000, "max_output": 16384, "capabilities": { "vision": true, "tool_use": true, "streaming": true } }, "claude-opus-4-7": { "display_name": "Claude Opus 4.7", "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'adaptive'}", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking" } }, "claude-haiku-4-5-20251001": { "display_name": "Claude Haiku 4.5", "tier": "fast", "context_window": 200000, "max_output": 64000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Pinned snapshot; same semantics as claude-haiku-4-5." } }, "claude-sonnet-4-5-20250929": { "display_name": "Claude Sonnet 4.5", "tier": "balanced-legacy", "context_window": 200000, "max_output": 64000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'adaptive'}", "notes": "Pinned snapshot of claude-sonnet-4-5." } }, "claude-opus-4-5-20251101": { "display_name": "Claude Opus 4.5", "tier": "flagship-legacy", "context_window": 200000, "max_output": 64000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Pinned snapshot of claude-opus-4-5." } }, "claude-opus-4-1-20250805": { "display_name": "Claude Opus 4.1", "tier": "flagship-legacy", "context_window": 200000, "max_output": 32000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Legacy 4.1 opus snapshot; extended thinking via budget_tokens." } }, "claude-opus-4-1": { "display_name": "Claude Opus 4.1", "tier": "flagship-legacy", "context_window": 200000, "max_output": 32000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Legacy 4.1 opus snapshot; extended thinking via budget_tokens." } }, "claude-sonnet-4-20250514": { "display_name": "Claude Sonnet 4", "tier": "balanced-legacy", "context_window": 200000, "max_output": 64000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Pinned snapshot. Retires 2026-06-15." } }, "claude-opus-4-20250514": { "display_name": "Claude Opus 4", "tier": "flagship-legacy", "context_window": 200000, "max_output": 32000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Pinned snapshot. Retires 2026-06-15." } }, "claude-opus-4-8": { "display_name": "Claude Opus 4.8", "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking Same request surface as Opus 4.7; sampling params (temperature/top_p/top_k) rejected." } }, "claude-sonnet-5": { "display_name": "Claude Sonnet 5", "tier": "balanced", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking Adaptive runs by DEFAULT when thinking is omitted; non-default sampling params rejected. New tokenizer: ~30% more tokens for the same text vs Sonnet 4.6." } }, "claude-sonnet-5-5": { "display_name": "Claude Sonnet 5.5", "tier": "balanced", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking Adaptive runs by DEFAULT when thinking is omitted; non-default sampling params rejected. New tokenizer: ~30% more tokens for the same text vs Sonnet 4.6." } }, "claude-fable-5": { "display_name": "Claude Fable 5", "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "thinking", "echo_back": "required", "notes": "Thinking is ALWAYS on — explicit {type:'disabled'} returns 400; omit the thinking param (adaptive is implicit). Raw chain of thought never returned; display:'summarized' gives a readable summary. Sampling params rejected. Depth via output_config.effort (low..max). Requires 30-day data retention (400 under ZDR). https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5" } }, "claude-fable-5-1": { "display_name": "Claude Fable 5.1", "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "thinking", "echo_back": "required", "notes": "Thinking is ALWAYS on — explicit {type:'disabled'} returns 400; omit the thinking param (adaptive is implicit). Raw chain of thought never returned; display:'summarized' gives a readable summary. Sampling params rejected. Depth via output_config.effort (low..max). Requires 30-day data retention (400 under ZDR). https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5" } }, "claude-fable-5-2": { "display_name": "Claude Fable 5.2", "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "thinking", "echo_back": "required", "notes": "Thinking is ALWAYS on — explicit {type:'disabled'} returns 400; omit the thinking param (adaptive is implicit). Raw chain of thought never returned; display:'summarized' gives a readable summary. Sampling params rejected. Depth via output_config.effort (low..max). Requires 30-day data retention (400 under ZDR). https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5" } } }, "cache": { "supported": "explicit", "control": "breakpoints", "default_ttl_min": 5, "extended_ttl_min": 60, "extended_ttl_param": "cache_control.ttl=1h", "isolation": "per_workspace", "isolation_confidence": "explicit", "min_tokens": 1024, "max_breakpoints": 4, "observable": true, "observability_field": "usage.cache_read_input_tokens + usage.cache_creation_input_tokens", "pollution_risk": "safe", "notes": "Requires explicit cache_control breakpoints — Orbital's Anthropic adapter does NOT currently set them, so caching is effectively off. 1h TTL: 2.0x write cost, 0.1x read; amortizes after ~3 hits. Min tokens vary by model: opus-4-x/haiku-4-5=4096, sonnet-4-6=2048, sonnet-4-5/4=1024. Per-workspace isolation since 2026-02-05. https://platform.claude.com/docs/en/build-with-claude/prompt-caching" } }, "deepseek": { "display_name": "DeepSeek", "base_url": "https://api.deepseek.com", "console_url": "https://platform.deepseek.com/api_keys", "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "deepseek-v4-flash", "deepseek-v4-pro" ], "notes": "deepseek-v4-flash and deepseek-v4-pro are unified hybrid models — set thinking.type='enabled' for reasoning, omit/disable for general chat. CRITICAL: v4 models REQUIRE reasoning_content to be echoed back on subsequent turns (especially after tool calls) — missing field returns HTTP 400. Legacy deepseek-chat (non-thinking) and deepseek-reasoner (thinking) are aliases, deprecating 2026-07-24. deepseek-reasoner has the OPPOSITE rule: reasoning_content must NOT be sent back (also 400s). OpenAI-compatible. Text-only, no vision.", "pricing": { "deepseek-chat": { "input_per_1m": 0.28, "cached_input_per_1m": 0.028, "output_per_1m": 0.42 }, "deepseek-reasoner": { "input_per_1m": 0.55, "cached_input_per_1m": 0.14, "output_per_1m": 2.19 }, "_default": { "input_per_1m": 0.28, "cached_input_per_1m": 0.028, "output_per_1m": 0.42 }, "deepseek-v4-flash": { "input_per_1m": 0.14, "cached_input_per_1m": 0.0028, "output_per_1m": 0.28 }, "deepseek-v4-pro": { "input_per_1m": 0.435, "cached_input_per_1m": 0.003625, "output_per_1m": 0.87 } }, "models": { "deepseek-chat": { "display_name": "DeepSeek V3.2", "tier": "flagship", "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Legacy alias for v4-flash non-thinking mode; no reasoning emitted. Deprecating 2026-07-24. https://api-docs.deepseek.com/quick_start/pricing" } }, "deepseek-reasoner": { "display_name": "DeepSeek R1", "tier": "reasoning", "context_window": 128000, "max_output": 16384, "capabilities": { "vision": false, "tool_use": false, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "forbidden", "enable": "model_only", "notes": "LEGACY R1 — docs explicitly say 'if reasoning_content is included in input messages, the API will return a 400 error.' Strip before resend. Deprecating 2026-07-24. https://api-docs.deepseek.com/guides/reasoning_model" } }, "_default": { "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "deepseek-v4-flash": { "display_name": "DeepSeek V4 Flash", "tier": "hybrid", "context_window": 1000000, "max_output": 384000, "capabilities": { "vision": false, "tool_use": true, "streaming": true, "json_mode": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking defaults enabled. Echo-back required for tool-call turns; safest to always include reasoning_content (empty string when absent). https://api-docs.deepseek.com/guides/thinking_mode" } }, "deepseek-v4-pro": { "display_name": "DeepSeek V4 Pro", "tier": "hybrid", "context_window": 1000000, "max_output": 384000, "capabilities": { "vision": false, "tool_use": true, "streaming": true, "json_mode": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking defaults enabled. Docs: 'If your code does not correctly pass back reasoning_content, the API will return a 400 error.' Use thinking.type=disabled to opt out. https://api-docs.deepseek.com/guides/thinking_mode" } } }, "cache": { "supported": "implicit", "control": "automatic", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "per_account", "isolation_confidence": "explicit", "min_tokens": 64, "max_breakpoints": null, "observable": true, "observability_field": "usage.prompt_cache_hit_tokens + usage.prompt_cache_miss_tokens", "pollution_risk": "likely_safe", "notes": "Disk-based cache, multi-day inactivity TTL. 64-token prefix units. Best-effort, no 100% hit guarantee. Most refresh-friendly cache in the registry. https://api-docs.deepseek.com/guides/kv_cache" } }, "moonshot": { "display_name": "Moonshot (Kimi)", "base_url": "https://api.moonshot.ai/v1", "china_base_url": "https://api.moonshot.cn/v1", "console_url": "https://platform.moonshot.cn/console/api-keys", "currency": "CNY", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "kimi-k3", "kimi-k2.6", "kimi-k2.5" ], "notes": "Hybrid k2.5/k2.6 models accept thinking={type:enabled|disabled}; thinking defaults enabled on k2.6. K2.6 also exposes a 'keep' param: keep='all' REQUIRES historical reasoning_content preserved verbatim (echo-back becomes mandatory). Dedicated kimi-k2-thinking / -turbo are always-on and require echo-back unconditionally. api.moonshot.cn (China) and api.moonshot.ai / platform.kimi.ai (global) are functionally identical.", "pricing": { "kimi-k2.5": { "input_per_1m": 0.6, "output_per_1m": 3.0 }, "kimi-k2-instruct": { "input_per_1m": 0.5, "output_per_1m": 2.4 }, "moonshot-v1-128k": { "input_per_1m": 2.0, "output_per_1m": 5.0 }, "moonshot-v1-32k": { "input_per_1m": 1.0, "output_per_1m": 3.0 }, "moonshot-v1-8k": { "input_per_1m": 0.2, "output_per_1m": 2.0 }, "_default": { "input_per_1m": 1.0, "output_per_1m": 3.0 }, "kimi-k3": { "input_per_1m": 3.0, "output_per_1m": 15.0 }, "kimi-k3.1-pro": { "input_per_1m": 2.0, "output_per_1m": 10.0 }, "kimi-k3.1-pro-highspeed": { "input_per_1m": 4.0, "output_per_1m": 20.0 } }, "models": { "kimi-k2.5": { "display_name": "Kimi K2.5", "tier": "balanced", "context_window": 262144, "max_output": null, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid via thinking={type:enabled|disabled}; no 'keep' param so historical reasoning is not retained — echo-back not enforced. https://platform.kimi.ai/docs/api/chat" } }, "kimi-k2-instruct": { "display_name": "Kimi K2 Instruct", "tier": "balanced", "context_window": 131072, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Instruct (non-thinking) variant." } }, "moonshot-v1-128k": { "display_name": "Moonshot v1 128K", "tier": "balanced", "context_window": 131072, "max_output": null, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "v1 family has no thinking parameter; emits no reasoning_content. https://platform.kimi.ai/docs/api/chat" } }, "moonshot-v1-32k": { "display_name": "Moonshot v1 32K", "tier": "fast", "context_window": 32768, "max_output": null, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "v1 family has no thinking parameter." } }, "_default": { "context_window": 131072, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "kimi-k2.6": { "display_name": "Kimi K2.6", "tier": "flagship", "context_window": 262144, "max_output": null, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Defaults to enabled. Adds 'keep' param: keep='all' upgrades echo_back to required (preserve historical reasoning_content verbatim). https://platform.kimi.ai/docs/api/chat" } }, "kimi-k2-turbo-preview": { "display_name": "Kimi K2 Turbo", "tier": "fast", "context_window": 262144, "max_output": null, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Turbo preview of K2; same hybrid semantics as kimi-k2.6.", "needs_verification": true } }, "kimi-k3": { "display_name": "Kimi K3", "tier": "flagship", "context_window": 1048576, "max_output": 131072, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "reasoning_content", "echo_back": "optional", "notes": "Reasoning always on — no toggle (reasoning_effort supports only 'max'; do NOT send the K2.x thinking param). temperature fixed at 1.0 — omit it. Vision input rejects public image URLs (base64 or ms:// only). https://platform.kimi.ai/docs/guide/kimi-k3-quickstart" } }, "kimi-k3.1-pro": { "display_name": "Kimi K3.1 Pro", "tier": "flagship", "context_window": 1048576, "max_output": 131072, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "reasoning_content", "echo_back": "optional", "notes": "Reasoning always on (efforts low/high/max, no 'none'). Same request surface as Kimi K3. Window from models.dev snapshot 2026-11-18.", "needs_verification": true } }, "kimi-k3.1-pro-highspeed": { "display_name": "Kimi K3.1 Pro Highspeed", "tier": "flagship", "context_window": 1048576, "max_output": 131072, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "reasoning_content", "echo_back": "optional", "notes": "Reasoning always on (efforts low/high/max, no 'none'). Same request surface as Kimi K3. Window from models.dev snapshot 2026-11-18.", "needs_verification": true } } }, "cache": { "supported": "both", "control": "resource", "default_ttl_min": 60, "extended_ttl_min": null, "extended_ttl_param": "ttl_caller_set + reset_ttl", "isolation": "per_account", "isolation_confidence": "implied", "min_tokens": null, "max_breakpoints": null, "observable": true, "observability_field": "usage.cached_tokens", "pollution_risk": "safe", "notes": "moonshot-v1 = explicit POST /v1/caching resource. kimi-k2.x = automatic prefix. Per-minute storage fee on explicit (~Y10/M tokens/min). Break-even cited at ~11 calls/hour. https://platform.moonshot.cn/blog/posts/context-caching" } }, "zhipu": { "display_name": "Zhipu AI (GLM)", "base_url": "https://api.z.ai/v1", "china_base_url": "https://open.bigmodel.cn/api/paas/v4", "console_url": "https://open.bigmodel.cn/usercenter/apikeys", "currency": "CNY", "supports_model_list": false, "sdk": "openai", "suggested_models": [ "glm-5.3", "glm-5.3-flash", "glm-5.2", "glm-5.1", "glm-5", "glm-4.7", "glm-4.6", "glm-4.5-air" ], "notes": "GLM-5.3 is latest flagship (1M ctx, reasoning always on); GLM-5.2 also 1M; GLM-5.1 200K. GLM-5/Turbo flagship tier. GLM-4.7 is the dedicated thinking model. All GLM 4.x/5.x are hybrid (thinking.type=enabled to turn on); echo-back is OPTIONAL — server respects thinking.clear_thinking flag (default true = strip history). Endpoint is OpenAI-compat at https://api.z.ai/api/paas/v4 (note: /paas/v4, NOT /v1). Vision served by separate models (glm-5v-turbo, glm-4.6v) not listed.", "pricing": { "glm-5": { "input_per_1m": 1.0, "output_per_1m": 3.2 }, "glm-4.7": { "input_per_1m": 0.6, "output_per_1m": 2.2 }, "glm-4.6": { "input_per_1m": 0.6, "output_per_1m": 2.2 }, "glm-4.6v": { "input_per_1m": 0.3, "output_per_1m": 0.9 }, "glm-4.6v-flash": { "input_per_1m": 0.0, "output_per_1m": 0.0 }, "glm-4.5": { "input_per_1m": 0.6, "output_per_1m": 2.2 }, "glm-4.5v": { "input_per_1m": 0.6, "output_per_1m": 1.8 }, "glm-4.5-air": { "input_per_1m": 0.2, "output_per_1m": 1.1 }, "_default": { "input_per_1m": 1.0, "output_per_1m": 3.0 }, "glm-5.1": { "input_per_1m": 1.4, "output_per_1m": 4.4 }, "glm-5-turbo": { "input_per_1m": 1.2, "output_per_1m": 4.0 }, "glm-5v-turbo": { "input_per_1m": 1.2, "output_per_1m": 4.0 }, "glm-4.7-flashx": { "input_per_1m": 0.07, "output_per_1m": 0.4 }, "glm-4.7-flash": { "input_per_1m": 0.0, "output_per_1m": 0.0 }, "glm-4.5-flash": { "input_per_1m": 0.0, "output_per_1m": 0.0 }, "glm-5.2": { "input_per_1m": 1.4, "output_per_1m": 4.4 }, "glm-5.3": { "input_per_1m": 1.4, "output_per_1m": 4.4 }, "glm-5.3-flash": { "input_per_1m": 0.15, "output_per_1m": 0.5 }, "glm-5.3-flashx": { "input_per_1m": 0.37, "output_per_1m": 1.25 } }, "models": { "glm-5": { "display_name": "GLM-5", "tier": "flagship", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Same hybrid model as 4.x line; thinking defaults enabled. https://docs.z.ai/api-reference/llm/chat-completion" } }, "glm-4.7": { "display_name": "GLM-4.7", "tier": "reasoning", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid (not always-on). Docs: missing/truncated/rewritten/reordered blocks degrade performance but do not 400. https://docs.z.ai/guides/llm/glm-4.7" } }, "glm-4.6": { "display_name": "GLM-4.6", "tier": "balanced", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid. clear_thinking=false to retain history. https://docs.z.ai/guides/llm/glm-4.6" } }, "glm-4.6v": { "display_name": "GLM-4.6V", "tier": "balanced", "context_window": 128000, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Vision model of the GLM-4.6 generation; hybrid thinking assumed. Window, vision and price from models.dev snapshot 2026-11-18.", "needs_verification": true } }, "glm-4.6v-flash": { "display_name": "GLM-4.6V-Flash", "tier": "fast", "context_window": 128000, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Vision model of the GLM-4.6 generation; hybrid thinking assumed. Window, vision and price from models.dev snapshot 2026-11-18.", "needs_verification": true } }, "glm-4.5": { "display_name": "GLM-4.5", "tier": "balanced", "context_window": 131072, "max_output": 98304, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "GLM-4.5 generation; hybrid thinking assumed like GLM-4.6. Window, vision and price from models.dev snapshot 2026-11-18.", "needs_verification": true } }, "glm-4.5v": { "display_name": "GLM-4.5V", "tier": "balanced", "context_window": 64000, "max_output": 16384, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "GLM-4.5 generation; hybrid thinking assumed like GLM-4.6. Window, vision and price from models.dev snapshot 2026-11-18.", "needs_verification": true } }, "glm-4.5-air": { "display_name": "GLM-4.5 Air", "tier": "fast", "context_window": 128000, "max_output": 96000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid. Echo-back governed by thinking.clear_thinking flag (default true = strip). https://docs.z.ai/guides/llm/glm-4.5" } }, "_default": { "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "glm-5.1": { "display_name": "GLM-5.1", "tier": "flagship", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking default enabled. Streaming delta carries reasoning_content. https://docs.z.ai/guides/llm/glm-5.1" } }, "glm-5-turbo": { "display_name": "GLM-5-Turbo", "tier": "flagship", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Turbo variant of glm-5; same hybrid semantics.", "needs_verification": true } }, "glm-5v-turbo": { "display_name": "GLM-5V-Turbo", "tier": "flagship", "context_window": 200000, "max_output": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Vision variant of glm-5-turbo; hybrid thinking assumed. Window, vision and price from models.dev snapshot 2026-11-18.", "needs_verification": true } }, "glm-4.7-flashx": { "display_name": "GLM-4.7 FlashX", "tier": "fast", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "FlashX variant of glm-4.7; same hybrid semantics.", "needs_verification": true } }, "glm-4.7-flash": { "display_name": "GLM-4.7 Flash", "tier": "fast", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Fast variant of glm-4.7; same hybrid semantics.", "needs_verification": true } }, "glm-4.5-flash": { "display_name": "GLM-4.5 Flash", "tier": "fast", "context_window": 128000, "max_output": 96000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Flash variant; same hybrid semantics.", "needs_verification": true } }, "glm-5.2": { "display_name": "GLM-5.2", "tier": "flagship", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking default enabled. Streaming delta carries reasoning_content. https://docs.z.ai/guides/llm/glm-5.2" } }, "glm-5.3": { "display_name": "GLM-5.3", "tier": "flagship", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Reasoning always on: thinking.type accepts only enabled (reasoning_effort low/high/max). Streaming delta carries reasoning_content. https://docs.z.ai/guides/llm/glm-5.3" } }, "glm-5.3-flash": { "display_name": "GLM-5.3-Flash", "tier": "fast", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Reasoning always on: thinking.type accepts only enabled. Streaming delta carries reasoning_content. https://docs.z.ai/guides/vlm/glm-5.3-flash" } }, "glm-5.3-flashx": { "display_name": "GLM-5.3-FlashX", "tier": "fast", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Paid faster tier of GLM-5.3-Flash; assumed the same always-on reasoning. Window, vision and price from models.dev snapshot 2026-11-18.", "needs_verification": true } } }, "cache": { "supported": "implicit", "control": "automatic", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "unclear", "isolation_confidence": "unclear", "min_tokens": null, "max_breakpoints": null, "observable": true, "observability_field": "usage.prompt_tokens_details.cached_tokens", "pollution_risk": "at_risk", "notes": "Least-documented cache in the registry. TTL, eviction policy, isolation, hash granularity all undocumented. Format-sensitive (whitespace/ordering). Recommend empirical measurement before relying on cache cost models. https://docs.z.ai/guides/capabilities/cache" } }, "qwen": { "display_name": "Alibaba Cloud Qwen (DashScope)", "base_url": "https://dashscope-intl.aliyuncs.com/compatible-mode/v1", "china_base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1", "console_url": "https://bailian.console.aliyun.com", "currency": "USD", "supports_model_list": false, "sdk": "openai", "suggested_models": [ "qwen3-max", "qwen3.5-plus", "qwen-plus", "qwen3.5-flash", "qwen-flash", "qwen3-coder-plus" ], "notes": "Qwen3-Max is the current flagship (262K ctx, hybrid thinking). All Qwen3 hybrid models default thinking OFF; enable via extra_body.enable_thinking=true. Echo-back is OPTIONAL but governed by preserve_thinking flag. Multimodal lives in qwen3.5-plus / qwen3.5-flash. qwen-turbo deprecated; use qwen-flash. Qwen3-coder is non-thinking. OpenAI-compatible endpoint (dashscope-intl for international, dashscope for CN — same semantics).", "pricing": { "qwen3.5-max": { "input_per_1m": 2.0, "output_per_1m": 8.0 }, "qwen3.5-plus": { "input_per_1m": 0.8, "output_per_1m": 3.0 }, "qwen3-max": { "input_per_1m": 1.0, "output_per_1m": 3.0 }, "qwen-plus": { "input_per_1m": 0.5, "output_per_1m": 2.0 }, "qwen-turbo": { "input_per_1m": 0.2, "output_per_1m": 0.6 }, "qwen3-coder-plus": { "input_per_1m": 0.5, "output_per_1m": 2.0 }, "_default": { "input_per_1m": 1.0, "output_per_1m": 3.0 } }, "models": { "qwen3.5-max": { "display_name": "Qwen 3.5 Max", "tier": "flagship", "context_window": 262144, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Latest Qwen3.5 flagship; hybrid thinking-mode (default off). Same enable_thinking + preserve_thinking pattern.", "needs_verification": true } }, "qwen3-max": { "display_name": "Qwen3-Max", "tier": "flagship", "context_window": 262144, "max_output": 65536, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking DEFAULT OFF. Enable via extra_body.enable_thinking=true. preserve_thinking=true to retain history. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" } }, "qwen3.5-plus": { "display_name": "Qwen3.5-Plus", "tier": "balanced", "context_window": 1000000, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" } }, "qwen-plus": { "display_name": "Qwen-Plus", "tier": "balanced", "context_window": 1000000, "max_output": 32768, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" } }, "qwen-turbo": { "display_name": "Qwen Turbo", "tier": "fast", "context_window": 131072, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Deprecated. Use qwen-flash." } }, "qwen3-coder-plus": { "display_name": "Qwen3-Coder-Plus", "tier": "code", "context_window": 1000000, "max_output": 65536, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Coder model; not listed among thinking-supported. https://help.aliyun.com/zh/model-studio/qwen-coder" } }, "_default": { "context_window": 131072, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "qwen-max": { "display_name": "Qwen-Max", "tier": "flagship", "context_window": 32768, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off.", "needs_verification": true } }, "qwen-flash": { "display_name": "Qwen-Flash", "tier": "fast", "context_window": 1000000, "max_output": 32768, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off. Replaces deprecated qwen-turbo. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" } }, "qwen3.5-flash": { "display_name": "Qwen3.5-Flash", "tier": "fast", "context_window": 1000000, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" } }, "qwen3-coder-flash": { "display_name": "Qwen3-Coder-Flash", "tier": "code", "context_window": 1000000, "max_output": 65536, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Coder model variant; not thinking-supported.", "needs_verification": true } } }, "cache": { "supported": "both", "control": "breakpoints", "default_ttl_min": 5, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "per_account_per_model", "isolation_confidence": "explicit", "min_tokens": 1024, "max_breakpoints": null, "observable": true, "observability_field": "usage.cached_tokens + usage.cache_creation_input_tokens", "pollution_risk": "likely_safe", "notes": "Anthropic-compatible cache_control:ephemeral. 5min TTL is FIXED and SHORT — main risk for long-running sessions. Implicit min: 256 tokens, 0.2x read. Explicit min: 1024 tokens, 0.1x read, 1.25x write. Per-model isolation means swapping qwen3-max <-> qwen-plus busts cache. https://www.alibabacloud.com/help/en/model-studio/context-cache" } }, "xai": { "display_name": "xAI", "base_url": "https://api.x.ai/v1", "console_url": "https://console.x.ai", "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "grok-4.20", "grok-4-1-fast-reasoning", "grok-4-1-fast-non-reasoning", "grok-code-fast-1" ], "notes": "Grok 4.20 and grok-4-1-fast-reasoning are auto-reasoning (reasoning_effort returns 400). On /chat/completions xAI's SDK exposes reasoning_content in stream deltas, but the official contract for echo-back is on the Responses API (reasoning.encrypted_content). Echo-back is OPTIONAL on xAI — no upstream 400 enforcement.", "pricing": { "grok-4": { "input_per_1m": 3.0, "output_per_1m": 15.0 }, "grok-4-1-fast-reasoning": { "input_per_1m": 0.2, "output_per_1m": 0.5 }, "grok-4-1-fast-non-reasoning": { "input_per_1m": 0.2, "output_per_1m": 0.5 }, "grok-code-fast-1": { "input_per_1m": 0.2, "output_per_1m": 1.5 }, "grok-3": { "input_per_1m": 3.0, "output_per_1m": 15.0 }, "_default": { "input_per_1m": 3.0, "output_per_1m": 15.0 }, "grok-4.20": { "input_per_1m": 2.0, "output_per_1m": 6.0 }, "grok-4.3": { "input_per_1m": 1.25, "output_per_1m": 2.5 }, "grok-4.5": { "input_per_1m": 2.0, "output_per_1m": 6.0 }, "grok-4.6": { "input_per_1m": 2.0, "output_per_1m": 6.0 }, "grok-4.7": { "input_per_1m": 2.0, "output_per_1m": 6.0 } }, "models": { "grok-4": { "display_name": "Grok 4 (legacy)", "tier": "flagship", "context_window": 256000, "max_output": 256000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "auto", "notes": "Auto-reasoning. Same xAI semantics as grok-4.20.", "needs_verification": true } }, "grok-4-1-fast-reasoning": { "display_name": "Grok 4.1 Fast (Reasoning)", "tier": "fast", "context_window": 2000000, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "grok-4-1-fast variant with reasoning enabled. reasoning_effort not accepted (auto). Echo-back optional. https://docs.x.ai/developers/models/grok-4-1-fast-reasoning" } }, "grok-4-1-fast-non-reasoning": { "display_name": "Grok 4.1 Fast (Non-Reasoning)", "tier": "fast", "context_window": 2000000, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning variant of grok-4-1-fast. https://docs.x.ai/developers/models/grok-4-1-fast-reasoning" } }, "grok-code-fast-1": { "display_name": "Grok Code Fast 1", "tier": "code", "context_window": 256000, "max_output": 10000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "auto", "notes": "Marketed as a reasoning model for agentic coding; xAI reasoning guide does not detail it explicitly. Probe before relying on. https://x.ai/news/grok-code-fast-1", "needs_verification": true } }, "grok-3": { "display_name": "Grok 3 (legacy)", "tier": "flagship", "context_window": 131072, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model.", "needs_verification": true } }, "_default": { "context_window": 131072, "max_output": 16384, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "grok-4.20": { "display_name": "Grok 4.20", "tier": "flagship", "context_window": 2000000, "max_output": 2000000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "auto", "notes": "Reasons automatically; reasoning_effort returns 400. reasoning_content visible on stream deltas via xAI SDK. Encrypted reasoning round-trip is optional. https://docs.x.ai/docs/guides/reasoning" } }, "grok-4.3": { "display_name": "Grok 4.3", "tier": "flagship", "context_window": 1000000, "max_output": 30000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "auto", "notes": "Same xAI reasoning semantics as grok-4.20 assumed. Window and price from models.dev snapshot 2026-11-18 (base tier; above 200K input xAI doubles the rate).", "needs_verification": true } }, "grok-4.5": { "display_name": "Grok 4.5", "tier": "flagship", "context_window": 500000, "max_output": 500000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "auto", "notes": "Same xAI reasoning semantics as grok-4.20 assumed. Window and price from models.dev snapshot 2026-11-18 (base tier; above 200K input xAI doubles the rate).", "needs_verification": true } }, "grok-4.6": { "display_name": "Grok 4.6", "tier": "flagship", "context_window": 500000, "max_output": 500000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "auto", "notes": "Same xAI reasoning semantics as grok-4.20 assumed. Window and price from models.dev snapshot 2026-11-18 (base tier; above 200K input xAI doubles the rate).", "needs_verification": true } }, "grok-4.7": { "display_name": "Grok 4.7", "tier": "flagship", "context_window": 500000, "max_output": 500000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "auto", "notes": "Same xAI reasoning semantics as grok-4.20 assumed. Window and price from models.dev snapshot 2026-11-18 (base tier; above 200K input xAI doubles the rate).", "needs_verification": true } } }, "cache": { "supported": "implicit", "control": "automatic", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "unclear", "isolation_confidence": "unclear", "min_tokens": null, "max_breakpoints": null, "observable": true, "observability_field": "usage.cached_prompt_text_tokens (native) / usage.prompt_tokens_details.cached_tokens (compat)", "pollution_risk": "at_risk", "notes": "Docs explicitly state 'evicted at any time due to server load' and 'not 100% guaranteed'. Set x-grok-conv-id header for per-session routing stickiness (hint, not isolation guarantee). TTL, min tokens, exact pricing multiplier all undocumented. https://docs.x.ai/developers/advanced-api-usage/prompt-caching" } }, "google": { "display_name": "Google Gemini", "base_url": "https://generativelanguage.googleapis.com/v1beta/openai", "console_url": "https://aistudio.google.com/apikey", "no_china_endpoint": true, "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "gemini-3.1-pro-preview", "gemini-3-flash-preview", "gemini-2.5-flash" ], "notes": "Gemini 3 family (3.1-pro-preview, 3-flash-preview) STRICTLY enforces thought_signature echo-back on function-call/tool turns — missing signature returns 400. On the OpenAI-compat endpoint, signatures live in tool_calls[*].extra_content.google.thought_signature; thought summaries appear as reasoning_content when extra_body.google.thinking_config.include_thoughts=true. Gemini 2.5 is lenient (no enforcement, quality degrades). Endpoint: https://generativelanguage.googleapis.com/v1beta/openai (OpenAI-compat shim).", "pricing": { "gemini-3-pro-preview": { "input_per_1m": 2.0, "output_per_1m": 12.0 }, "gemini-3-flash-preview": { "input_per_1m": 0.5, "output_per_1m": 3.0 }, "gemini-2.5-flash": { "input_per_1m": 0.3, "output_per_1m": 2.5 }, "gemini-2.5-pro": { "input_per_1m": 1.25, "output_per_1m": 10.0 }, "gemini-2.0-flash": { "input_per_1m": 0.1, "output_per_1m": 0.4 }, "_default": { "input_per_1m": 1.0, "output_per_1m": 5.0 }, "gemini-3.1-pro-preview": { "input_per_1m": 2.0, "output_per_1m": 12.0 }, "gemini-2.5-flash-lite": { "input_per_1m": 0.1, "output_per_1m": 0.4 } }, "models": { "gemini-3-pro-preview": { "display_name": "Gemini 3 Pro", "tier": "flagship", "context_window": 1000000, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:reasoning_effort=medium", "notes": "Same Gemini 3 strict signature validation.", "needs_verification": true } }, "gemini-3-flash-preview": { "display_name": "Gemini 3 Flash Preview", "tier": "fast", "context_window": 1048576, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:reasoning_effort=medium", "notes": "Same Gemini 3 strict signature validation on tool-use turns. https://ai.google.dev/gemini-api/docs/thinking" } }, "gemini-2.5-flash": { "display_name": "Gemini 2.5 Flash", "tier": "fast", "context_window": 1048576, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:reasoning_effort=medium", "notes": "Uses thinking_budget (-1 for dynamic). Lenient. https://ai.google.dev/gemini-api/docs/thinking" } }, "gemini-2.5-pro": { "display_name": "Gemini 2.5 Pro", "tier": "flagship", "context_window": 1048576, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:reasoning_effort=medium", "notes": "2.5-series does not strictly enforce signature echo-back; quality degrades if omitted. https://ai.google.dev/gemini-api/docs/thinking" } }, "gemini-2.0-flash": { "display_name": "Gemini 2.0 Flash", "tier": "fast", "context_window": 1000000, "max_output": 8192, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Pre-thinking-mode generation." } }, "_default": { "context_window": 1000000, "max_output": 8192, "capabilities": { "vision": true, "tool_use": true, "streaming": true } }, "gemini-3.1-pro-preview": { "display_name": "Gemini 3.1 Pro Preview", "tier": "flagship", "context_window": 1048576, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:reasoning_effort=medium", "notes": "Gemini 3 STRICTLY enforces thought_signature echo-back on tool/function-call turns (extra_content.google.thought_signature). Pure text turns lenient. Thought summaries surface as reasoning_content when extra_body.google.thinking_config.include_thoughts=true. https://ai.google.dev/gemini-api/docs/thought-signatures" } }, "gemini-2.5-flash-lite": { "display_name": "Gemini 2.5 Flash-Lite", "tier": "fast", "context_window": 1048576, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:reasoning_effort=medium", "notes": "Lite variant; same lenient 2.5 semantics.", "needs_verification": true } } }, "cache": { "supported": "both", "control": "resource", "default_ttl_min": 60, "extended_ttl_min": null, "extended_ttl_param": "ttl_caller_set", "isolation": "unclear", "isolation_confidence": "unclear", "min_tokens": 1024, "max_breakpoints": null, "observable": true, "observability_field": "usage_metadata.cached_content_token_count", "pollution_risk": "likely_safe", "notes": "Implicit auto-cache on Gemini 2.5+. Explicit cachedContents resource with caller-set TTL (default 1h). Storage fee proportional to (tokens x TTL) — keeping a 30k cache alive for hours can outweigh savings if hit volume is low. Min: Gemini 3 Flash=1024, Pro=4096. OpenAI-compat shim accepts extra_body.cached_content. https://ai.google.dev/gemini-api/docs/caching" } }, "mistral": { "display_name": "Mistral", "base_url": "https://api.mistral.ai/v1", "console_url": "https://console.mistral.ai/api-keys", "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "mistral-large-latest", "mistral-medium-latest", "magistral-medium-latest", "mistral-small-latest", "codestral-latest", "devstral-2-25-12" ], "notes": "Mistral Large 3 / Medium 3.1 / Small 4 are non-reasoning (multimodal/text). Magistral Medium/Small 1.2 are reasoning models with NATIVE chain-of-thought. Magistral-2507/2509 emit structured `type=thinking` chunks; Magistral-2506 used inline ... tags in content. Echo-back is not enforced (optional). Codestral = code+FIM. Devstral = agentic SWE. OpenAI-compatible at /v1/chat/completions.", "pricing": { "mistral-large-latest": { "input_per_1m": 0.5, "output_per_1m": 1.5 }, "mistral-medium-latest": { "input_per_1m": 0.4, "output_per_1m": 2.0 }, "magistral-medium-latest": { "input_per_1m": 2.0, "output_per_1m": 5.0 }, "magistral-small-latest": { "input_per_1m": 0.5, "output_per_1m": 1.5 }, "mistral-small-latest": { "input_per_1m": 0.15, "output_per_1m": 0.6 }, "codestral-latest": { "input_per_1m": 0.3, "output_per_1m": 0.9 }, "devstral-small-latest": { "input_per_1m": 0.06, "output_per_1m": 0.18 }, "_default": { "input_per_1m": 0.5, "output_per_1m": 2.0 }, "devstral-2-25-12": { "input_per_1m": 0.4, "output_per_1m": 0.9 } }, "models": { "mistral-large-latest": { "display_name": "Mistral Large 3", "tier": "flagship", "context_window": 1048576, "max_output": 262144, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model. https://docs.mistral.ai/capabilities/reasoning" } }, "mistral-medium-latest": { "display_name": "Mistral Medium 3.1", "tier": "balanced", "context_window": 262144, "max_output": 262144, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model." } }, "magistral-medium-latest": { "display_name": "Magistral Medium 1.2", "tier": "reasoning", "context_window": 128000, "max_output": 40000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "optional", "enable": "param:reasoning_effort=high", "notes": "Magistral 2507/2509 emits structured chunks with type=thinking. 2506 used inline ... tags in content. Echo-back not enforced upstream. https://docs.mistral.ai/capabilities/reasoning" } }, "magistral-small-latest": { "display_name": "Magistral Small 1.2", "tier": "reasoning", "context_window": 128000, "max_output": 40000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "optional", "enable": "param:reasoning_effort=high", "notes": "Same shape as magistral-medium. Newer 2509/2507 structured chunks; 2506 inline tags. https://docs.mistral.ai/capabilities/reasoning" } }, "mistral-small-latest": { "display_name": "Mistral Small 4", "tier": "fast", "context_window": 262144, "max_output": 256000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model." } }, "codestral-latest": { "display_name": "Codestral", "tier": "code", "context_window": 256000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Code+FIM model; non-reasoning." } }, "devstral-small-latest": { "display_name": "Devstral Small", "tier": "fast", "context_window": 256000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Devstral agentic SWE; non-reasoning.", "needs_verification": true } }, "_default": { "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "devstral-2-25-12": { "display_name": "Devstral 2", "tier": "code", "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Devstral 2.x agentic SWE; non-reasoning.", "needs_verification": true } } }, "cache": { "supported": "none", "control": "none", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "none", "isolation_confidence": "n_a", "min_tokens": null, "max_breakpoints": null, "observable": false, "observability_field": null, "pollution_risk": "n_a", "notes": "No prompt caching documented as of 2026-04-29. No cache_control field, no cached-token pricing tier, no cached_tokens in usage. Independent audit (arXiv 2502.07776v1) confirmed no detectable caching. https://docs.mistral.ai/api" } }, "groq": { "display_name": "Groq", "base_url": "https://api.groq.com/openai/v1", "console_url": "https://console.groq.com/keys", "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "openai/gpt-oss-120b", "llama-3.3-70b-versatile", "moonshotai/kimi-k2-instruct-0905", "llama-3.1-8b-instant" ], "notes": "Ultra-fast inference for open-source models. OpenAI-compatible at /openai/v1. Reasoning models (gpt-oss-120b/20b, qwen3-32b, deepseek-r1-distill) emit reasoning in a `reasoning` field (not reasoning_content). TWO mutually-exclusive params: reasoning_format (parsed/raw/hidden — for qwen/deepseek-r1-distill) and include_reasoning + reasoning_effort (for gpt-oss). Echo-back is OPTIONAL (not enforced upstream). Deprecations: deepseek-r1-distill-llama-70b (2025-10-02), kimi-k2-instruct-0905 (2026-03-23). Use gpt-oss-120b instead.", "pricing": { "llama-4-scout-17b-16e-instruct": { "input_per_1m": 0.11, "output_per_1m": 0.34 }, "deepseek-r1-distill-llama-70b": { "input_per_1m": 0.75, "output_per_1m": 0.99 }, "_default": { "input_per_1m": 0.5, "output_per_1m": 1.0 }, "llama-3.1-8b-instant": { "input_per_1m": 0.05, "output_per_1m": 0.08 }, "llama-3.3-70b-versatile": { "input_per_1m": 0.59, "output_per_1m": 0.79 }, "openai/gpt-oss-20b": { "input_per_1m": 0.075, "output_per_1m": 0.3 }, "openai/gpt-oss-120b": { "input_per_1m": 0.15, "output_per_1m": 0.6 }, "qwen/qwen3-32b": { "input_per_1m": 0.29, "output_per_1m": 0.59 }, "moonshotai/kimi-k2-instruct-0905": { "input_per_1m": 1.0, "output_per_1m": 3.0 }, "meta-llama/llama-4-scout-17b-16e-instruct": { "input_per_1m": 0.11, "output_per_1m": 0.34 } }, "models": { "llama-4-scout-17b-16e-instruct": { "display_name": "Llama 4 Scout", "tier": "fast", "context_window": 128000, "max_output": 8192, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model.", "needs_verification": true } }, "deepseek-r1-distill-llama-70b": { "display_name": "DeepSeek R1 Distill 70B", "tier": "reasoning", "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": false, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "param:reasoning_format=parsed", "notes": "DEPRECATED 2025-10-02. Same reasoning_format semantics as qwen3-32b on Groq; Groq does NOT enforce DeepSeek-style echo-back. https://console.groq.com/docs/reasoning" } }, "_default": { "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "llama-3.1-8b-instant": { "display_name": "Llama 3.1 8B Instant", "tier": "fast", "context_window": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model." } }, "llama-3.3-70b-versatile": { "display_name": "Llama 3.3 70B Versatile", "tier": "flagship", "context_window": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model." } }, "openai/gpt-oss-20b": { "display_name": "GPT-OSS 20B", "tier": "reasoning", "context_window": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "param:reasoning_effort=medium", "notes": "Same as gpt-oss-120b. https://console.groq.com/docs/reasoning" } }, "openai/gpt-oss-120b": { "display_name": "GPT-OSS 120B", "tier": "reasoning", "context_window": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "param:reasoning_effort=medium", "notes": "Reasoning in `reasoning` field on assistant message. reasoning_format param NOT supported for gpt-oss; use include_reasoning. Default returns reasoning. https://console.groq.com/docs/reasoning" } }, "qwen/qwen3-32b": { "display_name": "Qwen3 32B", "tier": "reasoning", "context_window": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "param:reasoning_format=parsed", "notes": "Use reasoning_format=parsed for `reasoning` field, raw to keep inline in content, hidden to suppress. include_reasoning is mutually exclusive with reasoning_format. https://console.groq.com/docs/reasoning" } }, "moonshotai/kimi-k2-instruct-0905": { "display_name": "Kimi K2 Instruct (0905)", "tier": "flagship", "context_window": 262144, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Kimi K2 instruct (non-reasoning) on Groq. DEPRECATED 2026-03-23 in favor of gpt-oss-120b." } }, "meta-llama/llama-4-scout-17b-16e-instruct": { "display_name": "Llama 4 Scout 17B (16E)", "tier": "flagship", "context_window": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model.", "needs_verification": true } } }, "cache": { "supported": "implicit", "control": "automatic", "default_ttl_min": 120, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "per_org", "isolation_confidence": "explicit", "min_tokens": 1024, "max_breakpoints": null, "observable": true, "observability_field": "usage.cached_tokens", "pollution_risk": "likely_safe", "notes": "Volatile-memory only (never persisted). ~2hr idle TTL. Min varies 128-1024 by model. Cached tokens DO NOT count toward rate limits. 1.0x write, 0.5x read. Prefix matches stop at first divergence — middle edits invalidate. https://console.groq.com/docs/prompt-caching" } }, "together": { "display_name": "Together AI", "base_url": "https://api.together.xyz/v1", "console_url": "https://api.together.xyz/settings/api-keys", "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "moonshotai/Kimi-K2.6", "deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-R1", "Qwen/Qwen3.5-397B-A17B", "meta-llama/Llama-3.3-70B-Instruct-Turbo", "openai/gpt-oss-120b" ], "notes": "Together does NOT unify reasoning shapes across models: DeepSeek-R1 emits inline ... tags inside content; gpt-oss-120b emits a separate `reasoning` field on the assistant message. Per-model contract varies — check the model page on together.ai. Echo-back is generally optional on Together (no documented gateway-level enforcement).", "pricing": { "_default": { "input_per_1m": 0.5, "output_per_1m": 1.0 }, "moonshotai/Kimi-K2.6": { "input_per_1m": 1.2, "output_per_1m": 4.5 }, "deepseek-ai/DeepSeek-V4-Pro": { "input_per_1m": 2.1, "output_per_1m": 4.4 }, "deepseek-ai/DeepSeek-R1": { "input_per_1m": 3.0, "output_per_1m": 7.0 }, "Qwen/Qwen3.5-397B-A17B": { "input_per_1m": 0.6, "output_per_1m": 3.6 }, "meta-llama/Llama-3.3-70B-Instruct-Turbo": { "input_per_1m": 0.88, "output_per_1m": 0.88 }, "openai/gpt-oss-120b": { "input_per_1m": 0.15, "output_per_1m": 0.6 } }, "models": { "_default": { "display_name": "Default", "tier": "fast", "context_window": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "moonshotai/Kimi-K2.6": { "display_name": "Kimi K2.6", "tier": "flagship", "context_window": 262144, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Mirrors upstream Moonshot K2.6 semantics. Together-specific behavior unverified.", "needs_verification": true } }, "deepseek-ai/DeepSeek-V4-Pro": { "display_name": "DeepSeek V4 Pro", "tier": "flagship", "context_window": 512000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Upstream DeepSeek v4-pro requires reasoning_content echo-back. Together's gateway behavior (whether it surfaces the field or inlines) needs verification on the model page. Treat as required to be safe. https://www.together.ai/models/deepseek-v4-pro", "needs_verification": true } }, "deepseek-ai/DeepSeek-R1": { "display_name": "DeepSeek R1", "tier": "reasoning", "context_window": 163839, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Together returns reasoning INLINE via ... tags in `content` (no separate reasoning_content field). Old R1 contract did not require echo-back. https://docs.together.ai/docs/deepseek-r1" } }, "Qwen/Qwen3.5-397B-A17B": { "display_name": "Qwen 3.5 397B A17B", "tier": "flagship", "context_window": 262144, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Mirrors upstream Qwen3 hybrid semantics. Verify on Together's model card.", "needs_verification": true } }, "meta-llama/Llama-3.3-70B-Instruct-Turbo": { "display_name": "Llama 3.3 70B Instruct Turbo", "tier": "fast", "context_window": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "field": null, "echo_back": "none", "enable": "model_only", "notes": "Non-reasoning model." } }, "openai/gpt-oss-120b": { "display_name": "GPT-OSS 120B", "tier": "reasoning", "context_window": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "param:reasoning_effort=medium", "notes": "Together exposes a SEPARATE `reasoning` field on the assistant message. Example: { message: { role, content, reasoning } }. https://docs.together.ai/docs/gpt-oss" } } }, "cache": { "supported": "dedicated_only", "control": "automatic", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "per_endpoint", "isolation_confidence": "implied", "min_tokens": null, "max_breakpoints": null, "observable": false, "observability_field": null, "pollution_risk": "safe", "notes": "Caching ONLY on Dedicated Endpoints (single-tenant); serverless has no cache (audit-confirmed). Non-disableable on Dedicated post-Feb-2026. No observability fields. https://docs.together.ai/docs/dedicated-inference" } }, "openrouter": { "display_name": "OpenRouter", "base_url": "https://openrouter.ai/api/v1", "console_url": "https://openrouter.ai/settings/keys", "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "anthropic/claude-opus-4.7", "anthropic/claude-sonnet-4.6", "google/gemini-3-flash-preview", "moonshotai/kimi-k2.6", "deepseek/deepseek-v3.2", "x-ai/grok-4.1-fast", "minimax/minimax-m2.7", "stepfun/step-3.5-flash" ], "notes": "OpenRouter normalizes upstream reasoning into three aliases: `reasoning` (string), `reasoning_content` (input alias), and `reasoning_details` (structured array, required for Anthropic tool-use continuity). Anthropic models on OR REQUIRE reasoning_details echo-back across tool-use turns; DeepSeek models pass through upstream's strict echo-back contract. Use reasoning.max_tokens, reasoning.effort, or reasoning.enabled to enable. reasoning.exclude=true suppresses output.", "pricing": { "_default": { "input_per_1m": 2.0, "output_per_1m": 8.0 } }, "models": { "_default": { "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "anthropic/claude-opus-4.7": { "display_name": "Claude Opus 4.7", "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "required", "enable": "param:reasoning.max_tokens=N", "notes": "OR normalizes Anthropic thinking into `reasoning` + `reasoning_details`. reasoning_details MUST be passed back unmodified across tool-use turns. https://openrouter.ai/docs/guides/best-practices/reasoning-tokens" } }, "anthropic/claude-sonnet-4.6": { "display_name": "Claude Sonnet 4.6", "tier": "flagship", "context_window": 1000000, "max_output": 8192, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "required", "enable": "param:reasoning.max_tokens=N", "notes": "Same Anthropic semantics; reasoning_details preserved verbatim across tool-use turns." } }, "google/gemini-3-flash-preview": { "display_name": "Gemini 3 Flash Preview", "tier": "fast", "context_window": 1048576, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "param:reasoning.max_tokens=N", "notes": "Gemini thinking via OR's unified reasoning param. Upstream Gemini does not require echo-back for pure text." } }, "moonshotai/kimi-k2.6": { "display_name": "Kimi K2.6", "tier": "flagship", "context_window": 256000, "max_output": 8192, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "param:reasoning.enabled=true", "notes": "OR unifies Moonshot reasoning_content into `reasoning`. Echo-back follows upstream Moonshot (optional unless keep='all').", "needs_verification": true } }, "deepseek/deepseek-v3.2": { "display_name": "DeepSeek V3.2", "tier": "reasoning", "context_window": 131072, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "required", "enable": "param:reasoning.enabled=true", "notes": "OR normalizes DeepSeek reasoning_content into `reasoning` (also accepts reasoning_content alias). DeepSeek upstream enforces echo-back; OR passes 400 through. Pass back `message.reasoning` (string) OR `reasoning_details` (array) unmodified. https://openrouter.ai/docs/guides/best-practices/reasoning-tokens" } }, "x-ai/grok-4.1-fast": { "display_name": "Grok 4.1 Fast", "tier": "fast", "context_window": 2000000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "param:reasoning.effort=high", "notes": "Grok via OR. reasoning.effort accepts low/medium/high/xhigh/minimal. No upstream echo-back enforcement." } }, "minimax/minimax-m2.7": { "display_name": "MiniMax M2.7", "tier": "flagship", "context_window": 196608, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "param:reasoning.enabled=true", "notes": "MiniMax M-series; OR central reasoning page lacks specifics. Confirm via model page on openrouter.ai.", "needs_verification": true } }, "stepfun/step-3.5-flash": { "display_name": "Step 3.5 Flash", "tier": "fast", "context_window": 262144, "max_output": 8192, "capabilities": { "vision": false, "tool_use": false, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning", "echo_back": "optional", "enable": "auto", "notes": "StepFun reasoning support on OR not authoritatively documented.", "needs_verification": true } } }, "cache": { "supported": "passthrough", "control": "passthrough", "default_ttl_min": null, "extended_ttl_min": 60, "extended_ttl_param": "passthrough:anthropic.cache_control.ttl=1h", "isolation": "passthrough", "isolation_confidence": "implied", "min_tokens": null, "max_breakpoints": 4, "observable": true, "observability_field": "usage.prompt_tokens_details.cached_tokens + cache_write_tokens + cache_discount", "pollution_risk": "at_risk", "notes": "Pass-through to upstream provider; OR does not maintain its own cache. Provider sticky routing activates only when cache-read pricing beats normal input — bypassed by provider.order. Pin provider.order per session to prevent silent re-routing breaking cache continuity. https://openrouter.ai/docs/guides/best-practices/prompt-caching" } }, "minimax": { "display_name": "MiniMax", "base_url": "https://api.minimax.io/v1", "china_base_url": "https://api.minimaxi.com/v1", "console_url": "https://platform.minimaxi.com/user-center/basic-information/interface-key", "currency": "USD", "supports_model_list": false, "sdk": "openai", "suggested_models": [ "MiniMax-M3", "MiniMax-M2.7", "MiniMax-M2.5" ], "notes": "OpenAI-compatible Chat Completions. International keys use api.minimax.io; China-region keys MUST use api.minimaxi.com (credentials are region-specific and not interchangeable). M-series reasoning is emitted inline in content as …; Orbital separates it into reasoning_content via the InlineThinkSplitter (ReasoningInfo.field=null contract). Model-list endpoint unverified — UI falls back to suggested_models. Pricing below is an estimate; confirm on https://platform.minimax.io/docs.", "pricing": { "MiniMax-M3": { "input_per_1m": 0.3, "cached_input_per_1m": 0.06, "output_per_1m": 1.2 }, "MiniMax-M2.7": { "input_per_1m": 0.3, "cached_input_per_1m": 0.06, "output_per_1m": 1.2 }, "MiniMax-M2.5": { "input_per_1m": 0.3, "cached_input_per_1m": 0.06, "output_per_1m": 1.2 }, "_default": { "input_per_1m": 0.3, "cached_input_per_1m": 0.06, "output_per_1m": 1.2 } }, "models": { "MiniMax-M3": { "display_name": "MiniMax M3", "tier": "flagship", "context_window": 1000000, "max_output": null, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think: M3 emits reasoning in … within content; Orbital's InlineThinkSplitter (field=null contract, default tag 'think') peels it into reasoning_content. (A v0.6.1 inline_tag='mm:think' override was a misdiagnosis from a single malformed lone close tag — M3 normally uses ; reverted in v0.6.2.) https://platform.minimax.io/docs/api-reference/text-openai-api", "needs_verification": true } }, "MiniMax-M2.7": { "display_name": "MiniMax M2.7", "tier": "flagship", "context_window": 196608, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think: M2-series emits reasoning in … within content; Orbital's InlineThinkSplitter (field=null contract) peels it into reasoning_content. Confirm specifics on https://platform.minimax.io/docs.", "needs_verification": true } }, "MiniMax-M2.5": { "display_name": "MiniMax M2.5", "tier": "balanced", "context_window": 196608, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think: M2-series emits reasoning in … within content; Orbital's InlineThinkSplitter (field=null contract) peels it into reasoning_content. Confirm specifics on https://platform.minimax.io/docs.", "needs_verification": true } }, "_default": { "context_window": 1000000, "max_output": 8192, "capabilities": { "vision": true, "tool_use": true, "streaming": true } } }, "cache": { "supported": "unknown", "control": "unknown", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "unknown", "isolation_confidence": "unknown", "min_tokens": null, "max_breakpoints": null, "observable": null, "observability_field": null, "pollution_risk": "unknown", "notes": "MiniMax caching behavior for the OpenAI-compatible endpoint is unverified." } }, "hunyuan": { "display_name": "Tencent Hunyuan (HY)", "base_url": "https://tokenhub-intl.tencentcloudmaas.com/v1", "china_base_url": "https://tokenhub.tencentmaas.com/v1", "console_url": "https://console.cloud.tencent.com/tokenhub/apikey", "currency": "CNY", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "hy4-preview", "hy3" ], "notes": "Tencent Cloud TokenHub serves the HY (混元) chat models over the OpenAI chat-completions protocol (an Anthropic /messages surface also exists). Flagship: hy4-preview (770B MoE, 1M ctx, 64K out, thinking on by default at high effort). Previous GA: hy3 (295B MoE, 256K ctx, 128K out, open weights). Regions: tokenhub-intl.tencentcloudmaas.com (global) vs tokenhub.tencentmaas.com (mainland, keys from console.cloud.tencent.com/tokenhub/apikey) — keys are region-specific, same model ids on both. The older direct endpoint api.hunyuan.cloud.tencent.com/v1 serves the legacy hunyuan-turbos/-t1 family, not HY. hy3 pricing is tiered by input length (1.2/4 under 16K, 1.6/6.4 to 32K, 2/8 above); the base tier is recorded here.", "pricing": { "hy4-preview": { "input_per_1m": 6.0, "cached_input_per_1m": 0.3, "output_per_1m": 18.0 }, "hy3": { "input_per_1m": 1.2, "cached_input_per_1m": 0.4, "output_per_1m": 4.0 }, "_default": { "input_per_1m": 1.2, "cached_input_per_1m": 0.4, "output_per_1m": 4.0 } }, "models": { "hy4-preview": { "display_name": "HY4 Preview", "tier": "flagship", "context_window": 1000000, "max_output": 64000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "needs_verification": true, "notes": "Thinking on by default (high effort); thinking={type:enabled|disabled} and reasoning_effort=high|low per Tencent Cloud docs; reasoning_content in the stream. Echo-back tolerance not yet probed live." } }, "hy3": { "display_name": "HY3", "tier": "balanced", "context_window": 192000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "needs_verification": true, "notes": "Reasoning-enabled; same thinking/reasoning_effort contract as hy4-preview. Not yet probed live." } }, "_default": { "context_window": 262144, "max_output": 65536, "capabilities": { "vision": false, "tool_use": true, "streaming": true } } }, "cache": { "supported": "implicit", "control": "automatic", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "per_account", "isolation_confidence": "inferred", "min_tokens": null, "max_breakpoints": null, "observable": false, "observability_field": null, "pollution_risk": "likely_safe", "notes": "TokenHub publishes cache-hit input pricing (0.3 for hy4-preview, 0.4-0.8 tiered for hy3) but the usage field that reports hits is undocumented; treated as implicit and unobservable until probed." } }, "tokendance": { "display_name": "TokenDance (词元跳动)", "base_url": "https://tokendance.space/gateway/v1", "console_url": "https://tokendance.space/keys", "china_only": true, "currency": "CNY", "supports_model_list": true, "default_model": "deepseek-v4-flash", "sdk": "openai", "extra_headers": { "X-App-URL": "https://github.com/zqiren/Orbital" }, "suggested_models": [ "deepseek-v4-pro", "glm-5.2", "kimi-k2.6", "minimax-m3", "deepseek-v4-flash" ], "notes": "China-mainland LLM API router (Watcha/观猹-incubated 'Chinese OpenRouter', Spec 47): one prepaid-credit key for DeepSeek, GLM, Qwen, Kimi, MiniMax, Seed, StepFun etc. OpenAI-compatible at /gateway/v1 with NATIVE protocol passthrough (no translation) — each model keeps its upstream reasoning/echo-back contract. Anthropic Messages also available at /gateway/v1/messages (v1 integration uses the OpenAI path). Unauthenticated GET /gateway/v1/models lists the catalog incl. non-chat models (image/video/TTS) — the models proxy filters to chat protocols via supported_protocols. Per-model pricing is set by TokenDance and shown only in their console; the _default below is a placeholder estimate pending machine-readable pricing (Spec 47 R7) — correct via pricing-overrides.json. Attribution (per app-attribution.md, confirmed via 产品方后台 2026-09-03): X-App-URL is the ONLY request-dimension attribution header and must byte-match the registered App URL and the OAuth app_url; X-Site-URL is a legacy fallback read only when X-App-URL is absent, and no other header (incl. X-App-Name) participates — do not add them back.", "pricing": { "_default": { "input_per_1m": 2.0, "cached_input_per_1m": 0.4, "output_per_1m": 8.0 } }, "models": { "deepseek-v4-pro": { "display_name": "DeepSeek V4 Pro", "tier": "hybrid", "context_window": 1048576, "max_output": 384000, "capabilities": { "vision": false, "tool_use": true, "streaming": true, "json_mode": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Upstream DeepSeek V4 Pro contract via TokenDance native passthrough — reasoning_content echo-back required on tool-call turns (upstream 400s without it).", "needs_verification": true } }, "deepseek-v4-flash": { "display_name": "DeepSeek V4 Flash", "tier": "hybrid", "context_window": 1048576, "max_output": 384000, "capabilities": { "vision": false, "tool_use": true, "streaming": true, "json_mode": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Upstream DeepSeek V4 Flash contract via TokenDance native passthrough — same echo-back rule as v4-pro.", "needs_verification": true } }, "glm-5.2": { "display_name": "GLM-5.2", "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Upstream GLM hybrid semantics via TokenDance passthrough; thinking defaults enabled.", "needs_verification": true } }, "kimi-k2.6": { "display_name": "Kimi K2.6", "tier": "flagship", "context_window": 262144, "max_output": null, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Upstream Moonshot K2.6 hybrid semantics via TokenDance passthrough; keep='all' upgrades echo-back to required.", "needs_verification": true } }, "minimax-m3": { "display_name": "MiniMax M3", "tier": "flagship", "context_window": 1000000, "max_output": null, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think via TokenDance passthrough: M3 emits reasoning as … in content; InlineThinkSplitter peels it into reasoning_content. Reasoning locked on (model_only).", "needs_verification": true } }, "_default": { "context_window": 131072, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } } }, "cache": { "supported": "passthrough", "control": "passthrough", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "passthrough", "isolation_confidence": "unclear", "min_tokens": null, "max_breakpoints": null, "observable": true, "observability_field": "usage.prompt_tokens_details.cached_tokens", "pollution_risk": "unknown", "notes": "Router passthrough to upstream provider caches. Live probe 2026-08-08: /chat/completions usage carries prompt_tokens_details.cached_tokens (OpenAI shape). Cache-hit billing is TokenDance per-model pricing — unverified." } }, "custom": { "display_name": "Custom / Self-Hosted", "base_url": null, "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [], "notes": "User provides base_url manually. Works with any OpenAI-compatible endpoint (Ollama, vLLM, LM Studio, etc.)", "pricing": { "_default": { "input_per_1m": 3.0, "output_per_1m": 15.0 } }, "models": { "_default": { "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } } }, "cache": { "supported": "unknown", "control": "unknown", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "unknown", "isolation_confidence": "unknown", "min_tokens": null, "max_breakpoints": null, "observable": null, "observability_field": null, "pollution_risk": "unknown", "notes": "Behavior depends on the user-configured endpoint. No general statement possible." } }, "opencode-zen": { "display_name": "OpenCode Zen", "base_url": "https://opencode.ai/zen/v1", "console_url": "https://opencode.ai/auth", "currency": "USD", "supports_model_list": true, "sdk": "openai", "suggested_models": [ "claude-sonnet-5", "claude-opus-5", "deepseek-v4-pro", "kimi-k3", "glm-5.2", "minimax-m3", "deepseek-v4-flash-free", "big-pickle" ], "notes": "Pay-as-you-go credits (auto-reload configurable), sold at cost. Orbital reaches the /chat/completions family (open models + the free tier), the /messages family (all Claude models and Qwen) and the Responses family (GPT and Grok; catalogued: gpt-6.2-sol, gpt-6-luna, gpt-5.6-luna, grok-4.7, grok-4.6). Gemini is served over Google's protocol, which Orbital does not speak, so use the Google provider directly for it. The free models (ids ending -free, plus big-pickle) work with NO payment method on file — verified on the wire. Note that a paid model with no balance returns HTTP 401 carrying a CreditsError body: the key is valid, the workspace just needs a payment method. Same API key as OpenCode Go. https://opencode.ai/docs/zen/ A Go subscription does NOT fund Zen: with Go active but no Zen credits, Zen's paid models still answer CreditsError while the free models keep working (verified 2026-08-19). The balance flows the other way — Go can fall back to Zen credits past a cap.", "pricing": { "big-pickle": { "input_per_1m": 0.0, "output_per_1m": 0.0, "cached_input_per_1m": 0.0 }, "claude-fable-5": { "input_per_1m": 10.0, "output_per_1m": 50.0, "cached_input_per_1m": 1.0, "cache_write_per_1m": 12.5 }, "claude-fable-5-1": { "input_per_1m": 10.0, "output_per_1m": 50.0, "cached_input_per_1m": 0.25, "cache_write_per_1m": 12.5 }, "claude-fable-5-2": { "input_per_1m": 8.0, "output_per_1m": 40.0, "cached_input_per_1m": 0.2, "cache_write_per_1m": 10.0 }, "claude-haiku-4-5": { "input_per_1m": 1.0, "output_per_1m": 5.0, "cached_input_per_1m": 0.1, "cache_write_per_1m": 1.25 }, "claude-opus-4-5": { "input_per_1m": 5.0, "output_per_1m": 25.0, "cached_input_per_1m": 0.5, "cache_write_per_1m": 6.25 }, "claude-opus-4-6": { "input_per_1m": 5.0, "output_per_1m": 25.0, "cached_input_per_1m": 0.5, "cache_write_per_1m": 6.25 }, "claude-opus-4-7": { "input_per_1m": 5.0, "output_per_1m": 25.0, "cached_input_per_1m": 0.5, "cache_write_per_1m": 6.25 }, "claude-opus-4-8": { "input_per_1m": 5.0, "output_per_1m": 25.0, "cached_input_per_1m": 0.5, "cache_write_per_1m": 6.25 }, "claude-opus-5": { "input_per_1m": 5.0, "output_per_1m": 25.0, "cached_input_per_1m": 0.5, "cache_write_per_1m": 6.25 }, "claude-opus-5-5": { "input_per_1m": 4.0, "output_per_1m": 20.0, "cached_input_per_1m": 0.2, "cache_write_per_1m": 5.0 }, "claude-sonnet-4-5": { "input_per_1m": 3.0, "output_per_1m": 15.0, "cached_input_per_1m": 0.3, "cache_write_per_1m": 3.75 }, "claude-sonnet-4-6": { "input_per_1m": 3.0, "output_per_1m": 15.0, "cached_input_per_1m": 0.3, "cache_write_per_1m": 3.75 }, "claude-sonnet-5": { "input_per_1m": 2.0, "output_per_1m": 10.0, "cached_input_per_1m": 0.2, "cache_write_per_1m": 2.5 }, "claude-sonnet-5-5": { "input_per_1m": 2.0, "output_per_1m": 10.0, "cached_input_per_1m": 0.1, "cache_write_per_1m": 2.5 }, "deepseek-v4-flash": { "input_per_1m": 0.22, "output_per_1m": 0.66, "cached_input_per_1m": 0.007 }, "deepseek-v4-flash-free": { "input_per_1m": 0.0, "output_per_1m": 0.0, "cached_input_per_1m": 0.0 }, "deepseek-v4-pro": { "input_per_1m": 0.66, "output_per_1m": 1.98, "cached_input_per_1m": 0.022 }, "glm-5": { "input_per_1m": 1.0, "output_per_1m": 3.2, "cached_input_per_1m": 0.2 }, "glm-5.1": { "input_per_1m": 1.4, "output_per_1m": 4.4, "cached_input_per_1m": 0.26 }, "glm-5.2": { "input_per_1m": 1.4, "output_per_1m": 4.4, "cached_input_per_1m": 0.26 }, "glm-5.3": { "input_per_1m": 1.4, "output_per_1m": 4.4, "cached_input_per_1m": 0.26 }, "glm-5.3-flash": { "input_per_1m": 0.15, "output_per_1m": 0.5, "cached_input_per_1m": 0.03 }, "hy3-free": { "input_per_1m": 0.0, "output_per_1m": 0.0, "cached_input_per_1m": 0.0 }, "kimi-k2.5": { "input_per_1m": 0.6, "output_per_1m": 3.0, "cached_input_per_1m": 0.1 }, "kimi-k2.6": { "input_per_1m": 0.95, "output_per_1m": 4.0, "cached_input_per_1m": 0.16 }, "kimi-k2.7-code": { "input_per_1m": 0.95, "output_per_1m": 4.0, "cached_input_per_1m": 0.19 }, "kimi-k3": { "input_per_1m": 3.0, "output_per_1m": 15.0, "cached_input_per_1m": 0.3 }, "laguna-s-2.1-free": { "input_per_1m": 0.0, "output_per_1m": 0.0, "cached_input_per_1m": 0.0 }, "mimo-v2.5-free": { "input_per_1m": 0.0, "output_per_1m": 0.0, "cached_input_per_1m": 0.0 }, "minimax-m2.5": { "input_per_1m": 0.3, "output_per_1m": 1.2, "cached_input_per_1m": 0.06 }, "minimax-m2.7": { "input_per_1m": 0.3, "output_per_1m": 1.2, "cached_input_per_1m": 0.06 }, "minimax-m3": { "input_per_1m": 0.3, "output_per_1m": 1.2, "cached_input_per_1m": 0.06 }, "nemotron-3-ultra-free": { "input_per_1m": 0.0, "output_per_1m": 0.0, "cached_input_per_1m": 0.0 }, "nemotron-3.5-lightning-free": { "input_per_1m": 0.0, "output_per_1m": 0.0, "cached_input_per_1m": 0.0 }, "qwen3.5-plus": { "input_per_1m": 0.2, "output_per_1m": 1.2, "cached_input_per_1m": 0.02, "cache_write_per_1m": 0.25 }, "qwen3.6-plus": { "input_per_1m": 0.5, "output_per_1m": 3.0, "cached_input_per_1m": 0.05, "cache_write_per_1m": 0.625 }, "qwen3.7-max": { "input_per_1m": 2.5, "output_per_1m": 7.5, "cached_input_per_1m": 0.5, "cache_write_per_1m": 3.125 }, "qwen3.7-plus": { "input_per_1m": 0.4, "output_per_1m": 1.6, "cached_input_per_1m": 0.04, "cache_write_per_1m": 0.5 }, "_default": { "input_per_1m": 3.0, "output_per_1m": 15.0 }, "claude-haiku-5-5": { "input_per_1m": 0.1, "output_per_1m": 0.5, "cached_input_per_1m": 0.01, "cache_write_per_1m": 0.125 }, "gpt-6-luna": { "input_per_1m": 0.1, "output_per_1m": 0.5, "cached_input_per_1m": 0.01 }, "gpt-5.6-luna": { "input_per_1m": 0.2, "output_per_1m": 1.2, "cached_input_per_1m": 0.02 }, "grok-4.7": { "input_per_1m": 2.0, "output_per_1m": 6.0, "cached_input_per_1m": 0.5 }, "grok-4.6": { "input_per_1m": 2.0, "output_per_1m": 6.0, "cached_input_per_1m": 0.5 }, "gpt-6.2-sol": { "input_per_1m": 2.0, "output_per_1m": 10.0, "cached_input_per_1m": 0.1 } }, "models": { "big-pickle": { "display_name": "Big Pickle", "context_window": 160000, "max_output": 32000, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "claude-fable-5": { "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "thinking", "echo_back": "required", "notes": "Thinking is ALWAYS on — explicit {type:'disabled'} returns 400; omit the thinking param (adaptive is implicit). Raw chain of thought never returned; display:'summarized' gives a readable summary. Sampling params rejected. Depth via output_config.effort (low..max). Requires 30-day data retention (400 under ZDR). https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5" }, "display_name": "Claude Fable 5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-fable-5-1": { "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "thinking", "echo_back": "required", "notes": "Thinking is ALWAYS on — explicit {type:'disabled'} returns 400; omit the thinking param (adaptive is implicit). Raw chain of thought never returned; display:'summarized' gives a readable summary. Sampling params rejected. Depth via output_config.effort (low..max). Requires 30-day data retention (400 under ZDR). https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5" }, "display_name": "Claude Fable 5.1", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-fable-5-2": { "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "thinking", "echo_back": "required", "notes": "Thinking is ALWAYS on — explicit {type:'disabled'} returns 400; omit the thinking param (adaptive is implicit). Raw chain of thought never returned; display:'summarized' gives a readable summary. Sampling params rejected. Depth via output_config.effort (low..max). Requires 30-day data retention (400 under ZDR). https://platform.claude.com/docs/en/about-claude/models/introducing-claude-fable-5" }, "display_name": "Claude Fable 5.2", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-haiku-4-5": { "tier": "fast", "context_window": 200000, "max_output": 32000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Manual extended thinking with budget_tokens. Echo-back required during tool use. https://platform.claude.com/docs/en/build-with-claude/extended-thinking" }, "display_name": "Claude Haiku 4.5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-opus-4-5": { "tier": "flagship", "context_window": 200000, "max_output": 16384, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'enabled',budget_tokens:N}", "notes": "Legacy 4.5 opus extended thinking with budget_tokens." }, "display_name": "Claude Opus 4.5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-opus-4-6": { "tier": "flagship-legacy", "context_window": 1000000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'adaptive'}", "notes": "Legacy 4.6 opus with extended thinking." }, "display_name": "Claude Opus 4.6", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-opus-4-7": { "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'adaptive'}", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking" }, "display_name": "Claude Opus 4.7", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-opus-4-8": { "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking Same request surface as Opus 4.7; sampling params (temperature/top_p/top_k) rejected." }, "display_name": "Claude Opus 4.8", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-opus-5": { "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking Same request surface as Opus 4.7; sampling params (temperature/top_p/top_k) rejected." }, "display_name": "Claude Opus 5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-opus-5-5": { "tier": "flagship", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking Same request surface as Opus 4.7; sampling params (temperature/top_p/top_k) rejected." }, "display_name": "Claude Opus 5.5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-sonnet-4-5": { "tier": "balanced", "context_window": 200000, "max_output": 32000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'adaptive'}", "notes": "Legacy 4.5 sonnet with extended thinking; same echo-back rule." }, "display_name": "Claude Sonnet 4.5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-sonnet-4-6": { "tier": "balanced", "context_window": 1000000, "max_output": 64000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "thinking", "echo_back": "required", "enable": "param:thinking={type:'adaptive'}", "notes": "Adaptive (recommended) or manual budget_tokens (deprecated). Echo-back of thinking blocks required during tool use. https://platform.claude.com/docs/en/build-with-claude/extended-thinking" }, "display_name": "Claude Sonnet 4.6", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-sonnet-5": { "tier": "balanced", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking Adaptive runs by DEFAULT when thinking is omitted; non-default sampling params rejected. New tokenizer: ~30% more tokens for the same text vs Sonnet 4.6." }, "display_name": "Claude Sonnet 5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-sonnet-5-5": { "tier": "balanced", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking only — manual budget_tokens returns 400. Thinking blocks (with signature) must be echoed unchanged on tool-use continuations. https://platform.claude.com/docs/en/build-with-claude/extended-thinking Adaptive runs by DEFAULT when thinking is omitted; non-default sampling params rejected. New tokenizer: ~30% more tokens for the same text vs Sonnet 4.6." }, "display_name": "Claude Sonnet 5.5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "deepseek-v4-flash": { "tier": "hybrid", "context_window": 1000000, "max_output": 384000, "capabilities": { "vision": false, "tool_use": true, "streaming": true, "json_mode": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking defaults enabled. Echo-back required for tool-call turns; safest to always include reasoning_content (empty string when absent). https://api-docs.deepseek.com/guides/thinking_mode" }, "display_name": "DeepSeek V4 Flash" }, "deepseek-v4-flash-free": { "display_name": "DeepSeek V4 Flash Free", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking defaults enabled. Echo-back required for tool-call turns; safest to always include reasoning_content (empty string when absent). https://api-docs.deepseek.com/guides/thinking_mode" } }, "deepseek-v4-pro": { "tier": "hybrid", "context_window": 1000000, "max_output": 384000, "capabilities": { "vision": false, "tool_use": true, "streaming": true, "json_mode": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking defaults enabled. Docs: 'If your code does not correctly pass back reasoning_content, the API will return a 400 error.' Use thinking.type=disabled to opt out. https://api-docs.deepseek.com/guides/thinking_mode" }, "display_name": "DeepSeek V4 Pro" }, "glm-5": { "tier": "flagship", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Same hybrid model as 4.x line; thinking defaults enabled. https://docs.z.ai/api-reference/llm/chat-completion" }, "display_name": "GLM 5" }, "glm-5.1": { "tier": "flagship", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking default enabled. Streaming delta carries reasoning_content. https://docs.z.ai/guides/llm/glm-5.1" }, "display_name": "GLM 5.1" }, "glm-5.2": { "tier": "flagship", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking default enabled. Streaming delta carries reasoning_content. https://docs.z.ai/guides/llm/glm-5.2" }, "display_name": "GLM 5.2" }, "glm-5.3": { "tier": "flagship", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Reasoning always on: thinking.type accepts only enabled (reasoning_effort low/high/max). Streaming delta carries reasoning_content. https://docs.z.ai/guides/llm/glm-5.3" }, "display_name": "GLM 5.3" }, "glm-5.3-flash": { "tier": "fast", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Reasoning always on: thinking.type accepts only enabled. Streaming delta carries reasoning_content. https://docs.z.ai/guides/vlm/glm-5.3-flash" }, "display_name": "GLM 5.3 Flash" }, "hy3-free": { "display_name": "Hy3 Free", "context_window": 190000, "max_output": 64000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Verified live on hy3-free: streams a named reasoning_content delta and accepts prior assistant turns that omit it. usage carries completion_tokens_details.reasoning_tokens." } }, "kimi-k2.5": { "tier": "balanced", "context_window": 262144, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid via thinking={type:enabled|disabled}; no 'keep' param so historical reasoning is not retained — echo-back not enforced. https://platform.kimi.ai/docs/api/chat" }, "display_name": "Kimi K2.5" }, "kimi-k2.6": { "tier": "flagship", "context_window": 262144, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Defaults to enabled. Adds 'keep' param: keep='all' upgrades echo_back to required (preserve historical reasoning_content verbatim). https://platform.kimi.ai/docs/api/chat" }, "display_name": "Kimi K2.6" }, "kimi-k2.7-code": { "tier": "flagship", "context_window": 262144, "max_output": null, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Defaults to enabled. Adds 'keep' param: keep='all' upgrades echo_back to required (preserve historical reasoning_content verbatim). https://platform.kimi.ai/docs/api/chat" }, "display_name": "Kimi K2.7 Code" }, "kimi-k3": { "tier": "flagship", "context_window": 1048576, "max_output": 131072, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "reasoning_content", "echo_back": "optional", "notes": "Reasoning always on — no toggle (reasoning_effort supports only 'max'; do NOT send the K2.x thinking param). temperature fixed at 1.0 — omit it. Vision input rejects public image URLs (base64 or ms:// only). https://platform.kimi.ai/docs/guide/kimi-k3-quickstart" }, "display_name": "Kimi K3" }, "laguna-s-2.1-free": { "display_name": "Laguna S 2.1 Free", "context_window": 256000, "max_output": 32000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "notes": "No reasoning surfaced in a live tool-calling probe (2026-08-19) — neither a reasoning_content delta nor inline ." } }, "mimo-v2.5-free": { "display_name": "MiMo-V2.5 Free", "context_window": 200000, "max_output": 32000, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "minimax-m2.5": { "tier": "balanced", "context_window": 196608, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think: M2-series emits reasoning in … within content; Orbital's InlineThinkSplitter (field=null contract) peels it into reasoning_content. Confirm specifics on https://platform.minimax.io/docs.", "needs_verification": true }, "display_name": "MiniMax M2.5" }, "minimax-m2.7": { "tier": "flagship", "context_window": 196608, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think: M2-series emits reasoning in … within content; Orbital's InlineThinkSplitter (field=null contract) peels it into reasoning_content. Confirm specifics on https://platform.minimax.io/docs.", "needs_verification": true }, "display_name": "MiniMax M2.7" }, "minimax-m3": { "tier": "flagship", "context_window": 512000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think: M3 emits reasoning in … within content; Orbital's InlineThinkSplitter (field=null contract, default tag 'think') peels it into reasoning_content. (A v0.6.1 inline_tag='mm:think' override was a misdiagnosis from a single malformed lone close tag — M3 normally uses ; reverted in v0.6.2.) https://platform.minimax.io/docs/api-reference/text-openai-api", "needs_verification": true }, "display_name": "MiniMax M3" }, "nemotron-3-ultra-free": { "display_name": "Nemotron 3 Ultra Free", "context_window": 1000000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "notes": "No reasoning surfaced in a live tool-calling probe (2026-08-19) — neither a reasoning_content delta nor inline ." } }, "nemotron-3.5-lightning-free": { "display_name": "Nemotron 3.5 Lightning Free", "context_window": 262144, "max_output": 262144, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": false, "notes": "No reasoning surfaced in a live tool-calling probe (2026-08-19) — neither a reasoning_content delta nor inline ." } }, "qwen3.5-plus": { "tier": "balanced", "context_window": 262144, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" }, "display_name": "Qwen3.5 Plus", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "qwen3.6-plus": { "tier": "balanced", "context_window": 262144, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" }, "display_name": "Qwen3.6 Plus", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "qwen3.7-max": { "tier": "flagship", "context_window": 262144, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Latest Qwen3.5 flagship; hybrid thinking-mode (default off). Same enable_thinking + preserve_thinking pattern.", "needs_verification": true }, "display_name": "Qwen3.7 Max", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "qwen3.7-plus": { "tier": "balanced", "context_window": 1000000, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" }, "display_name": "Qwen3.7 Plus", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "claude-haiku-5-5": { "tier": "fast", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking, on by default; manual budget_tokens returns 400. Thinking blocks must be echoed unchanged on tool-use continuations. Priced for prompts up to 100K tokens ($0.50/$2.50 beyond). Served ONLY over /messages on both OpenCode tiers — verified on the wire 2026-10-10 (400 ModelProtocolUnsupported on /chat/completions). https://platform.claude.com/docs/en/about-claude/models" }, "display_name": "Claude Haiku 5.5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen" }, "gpt-6-luna": { "display_name": "GPT-6 Luna", "tier": "fast", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served ONLY over OpenAI's Responses API on OpenCode (400 ModelProtocolUnsupported on /chat/completions, verified 2026-10-10); Orbital routes it to /responses." }, "sdk": "openai-responses" }, "gpt-5.6-luna": { "display_name": "GPT-5.6 Luna", "tier": "fast", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served ONLY over OpenAI's Responses API on OpenCode (400 ModelProtocolUnsupported on /chat/completions, verified 2026-10-10); Orbital routes it to /responses. Answered 403 unsupported_country_region_territory from this region (2026-10-10)." }, "sdk": "openai-responses" }, "grok-4.7": { "display_name": "Grok 4.7", "tier": "flagship", "context_window": 500000, "max_output": 500000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh" ], "notes": "Served ONLY over OpenAI's Responses API on OpenCode (400 ModelProtocolUnsupported on /chat/completions, verified 2026-10-10); Orbital routes it to /responses." }, "sdk": "openai-responses" }, "grok-4.6": { "display_name": "Grok 4.6", "tier": "flagship", "context_window": 500000, "max_output": 500000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh" ], "notes": "Served ONLY over OpenAI's Responses API on OpenCode (400 ModelProtocolUnsupported on /chat/completions, verified 2026-10-10); Orbital routes it to /responses." }, "sdk": "openai-responses" }, "gpt-6.2-sol": { "display_name": "GPT-6.2 Sol", "tier": "flagship", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh", "max" ], "notes": "Served ONLY over OpenAI's Responses API on OpenCode (400 ModelProtocolUnsupported on /chat/completions, verified 2026-10-10); Orbital routes it to /responses. Rejects reasoning effort 'none'." }, "sdk": "openai-responses" } }, "cache": { "supported": "passthrough", "control": "passthrough", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "passthrough", "isolation_confidence": "unknown", "min_tokens": null, "max_breakpoints": null, "observable": true, "observability_field": "prompt_tokens_details.cached_tokens", "pollution_risk": "unknown", "notes": "Aggregator: caching behavior is the upstream provider's, passed through. Cached-read/write rates are published per model and are billed at the upstream discount. Verified on the wire: usage carries prompt_tokens_details.cached_tokens." } }, "opencode-go": { "display_name": "OpenCode Go", "base_url": "https://opencode.ai/zen/go/v1", "console_url": "https://opencode.ai/auth", "currency": "USD", "supports_model_list": true, "sdk": "openai", "session_header": "x-opencode-session", "session_scope": "project", "extra_headers": { "User-Agent": "Orbital/{version}" }, "suggested_models": [ "glm-5.2", "kimi-k3", "claude-haiku-5-5", "glm-5.3", "minimax-m3", "kimi-k2.7-code", "mimo-v2.5", "deepseek-v4-pro" ], "notes": "$10/month subscription ($5 first month) with dollar-capped usage: $12 per 5 hours, $30/week, $60/month. The per-token rates below are what meters against those caps — they are not incremental charges, so spend readings show cap burn, not a bill. Enable 'Use balance' in the console to fall back to your Zen credits past a cap. Same API key as Zen. minimax-* and qwen* are served over the Anthropic /messages protocol on this tier, while the same minimax ids are /chat/completions on Zen — Orbital resolves that per model. claude-haiku-5-5 is /messages as well. grok-4.6, grok-4.7, gpt-5.6-luna and gpt-6-luna are served ONLY over OpenAI's Responses API (verified on the wire 2026-10-10); Orbital routes them to /responses per model (gpt-5.6-luna additionally answered 403 unsupported_country_region_territory from this region). Live /models also lists longcat-2.0, longcat-2.5-preview-free, step-5-preview-free, space-bunny and omen-alpha (all /chat/completions, reached via the provider default; no vendor specs published, so uncatalogued) and muse-spark-* (403 region-gated). Since 2026-09 the Go gateway rejects any request without a stable per-conversation x-opencode-session header (400 MissingSessionID, both wire protocols) and asks clients to identify with their own User-Agent — hence session_header + extra_headers on this entry. The header is a routing key (same value = same upstream = warm prompt cache), so session_scope \"project\" makes trigger-started sessions send one stable project: value instead of a fresh session id (chat sessions and workers keep their session id). https://opencode.ai/docs/go/ NOTE: deepseek-v4-flash and deepseek-v4-pro answer 403 RegionError on this tier until the workspace explicitly opts into China hosting in the OpenCode console — verified 2026-08-19; every other Go model works without it.", "pricing": { "deepseek-v4-flash": { "input_per_1m": 0.22, "output_per_1m": 0.66, "cached_input_per_1m": 0.007 }, "deepseek-v4-pro": { "input_per_1m": 0.66, "output_per_1m": 1.98, "cached_input_per_1m": 0.022 }, "glm-5.1": { "input_per_1m": 1.4, "output_per_1m": 4.4, "cached_input_per_1m": 0.26 }, "glm-5.2": { "input_per_1m": 1.4, "output_per_1m": 4.4, "cached_input_per_1m": 0.26 }, "glm-5.3": { "input_per_1m": 1.4, "output_per_1m": 4.4, "cached_input_per_1m": 0.26 }, "hy3": { "input_per_1m": 0.14, "output_per_1m": 0.58, "cached_input_per_1m": 0.035 }, "kimi-k2.6": { "input_per_1m": 0.95, "output_per_1m": 4.0, "cached_input_per_1m": 0.16 }, "kimi-k2.7-code": { "input_per_1m": 0.95, "output_per_1m": 4.0, "cached_input_per_1m": 0.19 }, "kimi-k3": { "input_per_1m": 3.0, "output_per_1m": 15.0, "cached_input_per_1m": 0.3 }, "kimi-k3.1-pro": { "input_per_1m": 2.0, "output_per_1m": 10.0, "cached_input_per_1m": 0.2 }, "mimo-v2.5": { "input_per_1m": 0.14, "output_per_1m": 0.28, "cached_input_per_1m": 0.0028 }, "mimo-v2.5-pro": { "input_per_1m": 0.435, "output_per_1m": 0.87, "cached_input_per_1m": 0.003625 }, "minimax-m2.5": { "input_per_1m": 0.3, "output_per_1m": 1.2, "cached_input_per_1m": 0.06, "cache_write_per_1m": 0.375 }, "minimax-m2.7": { "input_per_1m": 0.3, "output_per_1m": 1.2, "cached_input_per_1m": 0.06, "cache_write_per_1m": 0.375 }, "minimax-m3": { "input_per_1m": 0.3, "output_per_1m": 1.2, "cached_input_per_1m": 0.06 }, "minimax-m3.1-flash": { "input_per_1m": 0.15, "output_per_1m": 0.6, "cached_input_per_1m": 0.015 }, "qwen3.6-plus": { "input_per_1m": 0.5, "output_per_1m": 3.0, "cached_input_per_1m": 0.05, "cache_write_per_1m": 0.625 }, "qwen3.7-max": { "input_per_1m": 2.5, "output_per_1m": 7.5, "cached_input_per_1m": 0.5, "cache_write_per_1m": 3.125 }, "qwen3.7-plus": { "input_per_1m": 0.4, "output_per_1m": 1.6, "cached_input_per_1m": 0.04, "cache_write_per_1m": 0.5 }, "qwen3.8-max": { "input_per_1m": 2.0, "output_per_1m": 6.0, "cached_input_per_1m": 0.25, "cache_write_per_1m": 2.5 }, "_default": { "input_per_1m": 1.4, "output_per_1m": 4.4 }, "claude-haiku-5-5": { "input_per_1m": 0.1, "output_per_1m": 0.5, "cached_input_per_1m": 0.01, "cache_write_per_1m": 0.125 }, "glm-5.3-flash": { "input_per_1m": 0.15, "output_per_1m": 0.5, "cached_input_per_1m": 0.03 }, "mimo-v2.6-flash": { "input_per_1m": 0.14, "output_per_1m": 0.28, "cached_input_per_1m": 0.0028 }, "mimo-v2.6-pro": { "input_per_1m": 0.435, "output_per_1m": 0.87, "cached_input_per_1m": 0.003625 }, "qwen3.8-flash": { "input_per_1m": 0.15, "output_per_1m": 0.47, "cached_input_per_1m": 0.016, "cache_write_per_1m": 0.2 }, "deepseek-v4.1-flash": { "input_per_1m": 0.15, "output_per_1m": 0.6, "cached_input_per_1m": 0.003 }, "hy4-preview": { "input_per_1m": 0.834, "output_per_1m": 2.501, "cached_input_per_1m": 0.042 }, "gpt-6-luna": { "input_per_1m": 0.1, "output_per_1m": 0.5, "cached_input_per_1m": 0.01 }, "gpt-5.6-luna": { "input_per_1m": 0.2, "output_per_1m": 1.2, "cached_input_per_1m": 0.02 }, "grok-4.7": { "input_per_1m": 2.0, "output_per_1m": 6.0, "cached_input_per_1m": 0.5 }, "grok-4.6": { "input_per_1m": 2.0, "output_per_1m": 6.0, "cached_input_per_1m": 0.5 } }, "models": { "deepseek-v4-flash": { "tier": "hybrid", "context_window": 1000000, "max_output": 384000, "capabilities": { "vision": false, "tool_use": true, "streaming": true, "json_mode": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking defaults enabled. Echo-back required for tool-call turns; safest to always include reasoning_content (empty string when absent). https://api-docs.deepseek.com/guides/thinking_mode" }, "display_name": "DeepSeek V4 Flash" }, "deepseek-v4-pro": { "tier": "hybrid", "context_window": 1000000, "max_output": 384000, "capabilities": { "vision": false, "tool_use": true, "streaming": true, "json_mode": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking defaults enabled. Docs: 'If your code does not correctly pass back reasoning_content, the API will return a 400 error.' Use thinking.type=disabled to opt out. https://api-docs.deepseek.com/guides/thinking_mode" }, "display_name": "DeepSeek V4 Pro" }, "glm-5.1": { "tier": "flagship", "context_window": 200000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking default enabled. Streaming delta carries reasoning_content. https://docs.z.ai/guides/llm/glm-5.1" }, "display_name": "GLM-5.1" }, "glm-5.2": { "tier": "flagship", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking default enabled. Streaming delta carries reasoning_content. https://docs.z.ai/guides/llm/glm-5.2" }, "display_name": "GLM-5.2" }, "glm-5.3": { "tier": "flagship", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Reasoning always on: thinking.type accepts only enabled (reasoning_effort low/high/max). Streaming delta carries reasoning_content. https://docs.z.ai/guides/llm/glm-5.3" }, "display_name": "GLM-5.3" }, "hy3": { "display_name": "Hy3", "context_window": 192000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Verified live on hy3-free: streams a named reasoning_content delta and accepts prior assistant turns that omit it. usage carries completion_tokens_details.reasoning_tokens." } }, "kimi-k2.6": { "tier": "flagship", "context_window": 262144, "max_output": null, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Defaults to enabled. Adds 'keep' param: keep='all' upgrades echo_back to required (preserve historical reasoning_content verbatim). https://platform.kimi.ai/docs/api/chat" }, "display_name": "Kimi K2.6" }, "kimi-k2.7-code": { "tier": "flagship", "context_window": 262144, "max_output": null, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "notes": "Defaults to enabled. Adds 'keep' param: keep='all' upgrades echo_back to required (preserve historical reasoning_content verbatim). https://platform.kimi.ai/docs/api/chat" }, "display_name": "Kimi K2.7 Code" }, "kimi-k3": { "tier": "flagship", "context_window": 1048576, "max_output": 131072, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "reasoning_content", "echo_back": "optional", "notes": "Reasoning always on — no toggle (reasoning_effort supports only 'max'; do NOT send the K2.x thinking param). temperature fixed at 1.0 — omit it. Vision input rejects public image URLs (base64 or ms:// only). https://platform.kimi.ai/docs/guide/kimi-k3-quickstart" }, "display_name": "Kimi K3" }, "kimi-k3.1-pro": { "tier": "flagship", "context_window": 1048576, "max_output": 131072, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "model_only", "field": "reasoning_content", "echo_back": "optional", "notes": "Reasoning always on (efforts low/high/max, no 'none'; OpenCode lists reasoning_content). Same request surface as Kimi K3. Window and vision from opencode.ai/zen/go/v1/models 2026-11-19.", "needs_verification": true }, "display_name": "Kimi K3.1 Pro" }, "mimo-v2.5": { "display_name": "MiMo-V2.5", "context_window": 1000000, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "mimo-v2.5-pro": { "display_name": "MiMo-V2.5-Pro", "context_window": 1048576, "max_output": 128000, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "minimax-m2.5": { "tier": "balanced", "context_window": 196608, "max_output": 32768, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think: M2-series emits reasoning in … within content; Orbital's InlineThinkSplitter (field=null contract) peels it into reasoning_content. Confirm specifics on https://platform.minimax.io/docs.", "needs_verification": true }, "display_name": "MiniMax M2.5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen/go" }, "minimax-m2.7": { "tier": "flagship", "context_window": 196608, "max_output": 32768, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think: M2-series emits reasoning in … within content; Orbital's InlineThinkSplitter (field=null contract) peels it into reasoning_content. Confirm specifics on https://platform.minimax.io/docs.", "needs_verification": true }, "display_name": "MiniMax M2.7", "sdk": "anthropic", "base_url": "https://opencode.ai/zen/go" }, "minimax-m3": { "tier": "flagship", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": null, "echo_back": "optional", "enable": "model_only", "notes": "Inline-think: M3 emits reasoning in … within content; Orbital's InlineThinkSplitter (field=null contract, default tag 'think') peels it into reasoning_content. (A v0.6.1 inline_tag='mm:think' override was a misdiagnosis from a single malformed lone close tag — M3 normally uses ; reverted in v0.6.2.) https://platform.minimax.io/docs/api-reference/text-openai-api", "needs_verification": true }, "display_name": "MiniMax M3", "sdk": "anthropic", "base_url": "https://opencode.ai/zen/go" }, "minimax-m3.1-flash": { "tier": "fast", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Served over Chat Completions (unlike minimax-m3 on Go, which is Anthropic messages); OpenCode lists reasoning_field reasoning_content and efforts without 'none'. opencode.ai/zen/go/v1/models 2026-11-19.", "needs_verification": true }, "display_name": "MiniMax M3.1 Flash" }, "qwen3.6-plus": { "tier": "balanced", "context_window": 1000000, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" }, "display_name": "Qwen3.6 Plus", "sdk": "anthropic", "base_url": "https://opencode.ai/zen/go" }, "qwen3.7-max": { "tier": "flagship", "context_window": 262144, "max_output": 32768, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Latest Qwen3.5 flagship; hybrid thinking-mode (default off). Same enable_thinking + preserve_thinking pattern.", "needs_verification": true }, "display_name": "Qwen3.7 Max", "sdk": "anthropic", "base_url": "https://opencode.ai/zen/go" }, "qwen3.7-plus": { "tier": "balanced", "context_window": 1000000, "max_output": 65536, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Hybrid; thinking default off. https://www.alibabacloud.com/help/en/model-studio/deep-thinking" }, "display_name": "Qwen3.7 Plus", "sdk": "anthropic", "base_url": "https://opencode.ai/zen/go" }, "qwen3.8-max": { "tier": "flagship", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Latest Qwen3.5 flagship; hybrid thinking-mode (default off). Same enable_thinking + preserve_thinking pattern.", "needs_verification": true }, "display_name": "Qwen3.8 Max", "sdk": "anthropic", "base_url": "https://opencode.ai/zen/go" }, "claude-haiku-5-5": { "tier": "fast", "context_window": 1000000, "max_output": 128000, "capabilities": { "streaming": true, "tool_use": true, "vision": true }, "reasoning": { "supported": true, "enable": "param:thinking={type:'adaptive'}", "field": "thinking", "echo_back": "required", "notes": "Adaptive thinking, on by default; manual budget_tokens returns 400. Thinking blocks must be echoed unchanged on tool-use continuations. Priced for prompts up to 100K tokens ($0.50/$2.50 beyond). Served ONLY over /messages on both OpenCode tiers — verified on the wire 2026-10-10 (400 ModelProtocolUnsupported on /chat/completions). https://platform.claude.com/docs/en/about-claude/models" }, "display_name": "Claude Haiku 5.5", "sdk": "anthropic", "base_url": "https://opencode.ai/zen/go" }, "glm-5.3-flash": { "tier": "fast", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "model_only", "notes": "Reasoning always on: thinking.type accepts only enabled. Streaming delta carries reasoning_content. https://docs.z.ai/guides/vlm/glm-5.3-flash" }, "display_name": "GLM-5.3-Flash" }, "mimo-v2.6-flash": { "display_name": "MiMo-V2.6-Flash", "context_window": 1048576, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "mimo-v2.6-pro": { "display_name": "MiMo-V2.6-Pro", "context_window": 1048576, "max_output": 131072, "capabilities": { "vision": false, "tool_use": true, "streaming": true } }, "qwen3.8-flash": { "tier": "fast", "context_window": 1000000, "max_output": 131072, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:enable_thinking=true", "notes": "Latest Qwen3.5 flagship; hybrid thinking-mode (default off). Same enable_thinking + preserve_thinking pattern.", "needs_verification": true }, "display_name": "Qwen3.8 Flash" }, "deepseek-v4.1-flash": { "tier": "hybrid", "context_window": 1000000, "max_output": 384000, "capabilities": { "vision": false, "tool_use": true, "streaming": true, "json_mode": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "required", "enable": "param:thinking.type=enabled", "notes": "Hybrid; thinking defaults enabled. Echo-back required for tool-call turns; safest to always include reasoning_content (empty string when absent). https://api-docs.deepseek.com/guides/thinking_mode" }, "display_name": "DeepSeek V4.1 Flash" }, "hy4-preview": { "display_name": "HY4 Preview", "tier": "flagship", "context_window": 1000000, "max_output": 64000, "capabilities": { "vision": false, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "optional", "enable": "param:thinking.type=enabled", "needs_verification": true, "notes": "Thinking on by default (high effort); thinking={type:enabled|disabled} and reasoning_effort=high|low per Tencent Cloud docs; reasoning_content in the stream. Echo-back tolerance not yet probed live." } }, "gpt-6-luna": { "display_name": "GPT-6 Luna", "tier": "fast", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served ONLY over OpenAI's Responses API on OpenCode (400 ModelProtocolUnsupported on /chat/completions, verified 2026-10-10); Orbital routes it to /responses." }, "sdk": "openai-responses" }, "gpt-5.6-luna": { "display_name": "GPT-5.6 Luna", "tier": "fast", "context_window": 922000, "max_output": 128000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "none", "low", "medium", "high", "xhigh", "max" ], "notes": "Served ONLY over OpenAI's Responses API on OpenCode (400 ModelProtocolUnsupported on /chat/completions, verified 2026-10-10); Orbital routes it to /responses. Answered 403 unsupported_country_region_territory from this region (2026-10-10)." }, "sdk": "openai-responses" }, "grok-4.7": { "display_name": "Grok 4.7", "tier": "flagship", "context_window": 500000, "max_output": 500000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh" ], "notes": "Served ONLY over OpenAI's Responses API on OpenCode (400 ModelProtocolUnsupported on /chat/completions, verified 2026-10-10); Orbital routes it to /responses." }, "sdk": "openai-responses" }, "grok-4.6": { "display_name": "Grok 4.6", "tier": "flagship", "context_window": 500000, "max_output": 500000, "capabilities": { "vision": true, "tool_use": true, "streaming": true }, "reasoning": { "supported": true, "field": "reasoning_content", "echo_back": "none", "enable": "param:reasoning_effort=medium", "efforts": [ "low", "medium", "high", "xhigh" ], "notes": "Served ONLY over OpenAI's Responses API on OpenCode (400 ModelProtocolUnsupported on /chat/completions, verified 2026-10-10); Orbital routes it to /responses." }, "sdk": "openai-responses" } }, "cache": { "supported": "passthrough", "control": "passthrough", "default_ttl_min": null, "extended_ttl_min": null, "extended_ttl_param": null, "isolation": "passthrough", "isolation_confidence": "unknown", "min_tokens": null, "max_breakpoints": null, "observable": true, "observability_field": "prompt_tokens_details.cached_tokens", "pollution_risk": "unknown", "notes": "Aggregator: caching behavior is the upstream provider's, passed through. Cached-read/write rates are published per model and are billed at the upstream discount. Verified on the wire: usage carries prompt_tokens_details.cached_tokens." } } }, "defaults": { "unknown_model": { "context_window": 128000, "max_output": 8192, "capabilities": { "vision": false, "tool_use": true, "streaming": true } } } }