# Model: https://docs.mistral.ai/models/mistral-small-4-0-26-03 (API IDs mistral-small-2603, mistral-small-latest; $0.15 / $0.6). # Default: reasoning is off when reasoning_effort is omitted (live: no thinking chunk, 0 reasoning tokens). # Mistral API (api.mistral.ai), docs and live checks 2026-10-19: # - Pricing: https://docs.mistral.ai/inference/pricing (cached input is billed at 10% of input). # - Prompt caching: https://docs.mistral.ai/studio/conversations/advanced/prompt-caching (automatic prefix # cache in 64-token blocks, no write price, optional prompt_cache_key). Usage is reported when streaming. # - Tools: https://docs.mistral.ai/api/endpoint/chat (tool_choice auto|none|any|required or a named # function; function.strict). All five tool_choice forms worked live. # - Structured output: https://docs.mistral.ai/studio/conversations/structured-output (json_schema # response_format returned schema-valid JSON live). # - PDF: chat takes document_url PDFs (https://docs.mistral.ai/studio/document-processing/document_qna); # a base64 PDF was read correctly live. # - Sampling: temperature and top_p; no top_k (live 400 "top_k sampling is not enabled for this model"). # - Mid-conversation (live): a system message after the first assistant turn is accepted in place, # keeps the cached prefix and is obeyed. Adding or removing tools between requests keeps the cached # prefix (Mistral renders tools after it). # - Reasoning: https://docs.mistral.ai/studio/conversations/reasoning. Effort: reasoning_effort = none|high # (other values return 400 with "supported values: high, none"). Switching effort between turns keeps # the cached prefix (live). schema = 2 name = "Mistral Small 4" description = "Fast Mistral production model for chat, extraction, and cost-sensitive agents" family = "mistral-small" open_weights = true knowledge = "2025-06" release_date = "2026-03-16" [modalities] input = ["text", "image/png", "image/jpeg", "image/webp", "image/gif", "application/pdf"] output = ["text"] [capabilities] tools = { supported = true, strict = true, choice = ["auto", "none", "required", "tool"], updates = true } reasoning = { supported = true, default = { type = "effort", value = "none" }, effort = ["none", "high"] } structured_output = true temperature = true top_p = true top_k = false system_updates = true effort_updates = true stream_usage = true prompt_cache_key = true cache = { supported = true, implicit = { min_tokens = 64 }, explicit = false } [limit] context = 262_144 output = 256_000 [cost.text.input] price = 0.15 cache = { read = 0.015 } [cost.text.output] price = 0.6