--- name: gemini description: "Google Gemini chat models via inference.sh CLI. Models: Gemini 4 Argon (flagship), Gemini 3.8 Flash (most capable Flash), Gemini 3.5 Flash-Lite (fastest, lowest cost). 1M token context, thinking levels, vision, file input, tool use, structured output. Runs on Vertex AI. Use for: coding, autonomous agents, long documents, finance, law, science, high-throughput pipelines, image understanding, chat. Triggers: gemini, google gemini, gemini api, gemini 4, gemini 4 argon, argon, gemini flash, gemini 3.8 flash, gemini flash lite, gemini 3.5 flash-lite, vertex ai, google llm, google ai model, gemini chat" allowed-tools: Bash(belt *) --- > **Install the belt CLI skill:** `npx skills add belt-sh/cli` # Gemini Run Google's Gemini chat models via [inference.sh](https://inference.sh) CLI. ## Quick Start > Requires inference.sh CLI (`belt`). [Install instructions](https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md) ```bash belt login belt app run google/gemini-3-8-flash --input '{"text": "Explain how HTTPS certificate validation works"}' ``` ## Models | Model | App ID | Best For | |-------|--------|----------| | Gemini 4 Argon | `google/gemini-4-argon` | Flagship: deep multi-step work in coding, finance, law, security, science | | Gemini 3.8 Flash | `google/gemini-3-8-flash` | Most capable Flash: long-horizon software engineering, autonomous agents, enterprise workflows | | Gemini 3.5 Flash-Lite | `google/gemini-3-5-flash-lite` | Fastest, most cost-effective 3.5-class model (350 output tokens/s), high-throughput agentic workflows | All run via Vertex AI with vision and tool use. Argon and 3.8 Flash have a 1M token context window; Argon supports thinking levels low to high. Other Gemini apps (older chat versions, image models, and the Gemini Omni video models): `belt app search gemini`. `google/gemini-omni-1-2-flash` is a video generation model, not a chat model. ## Inputs | Field | Description | |-------|-------------| | `text` | The user message | | `system_prompt` | System instructions | | `images` | Image files or URLs (vision) | | `files` | Document files | | `context` | Prior messages: `[{"role": "user" \| "assistant", "text": "..."}]` | | `reasoning_effort` | `none` (default), `low`, `medium`, `high` | | `temperature` | Default 1.0 | | `top_k` | Default 40, -1 to disable | | `max_tokens` | Default 64000 | | `tools` | Function tool definitions | | `tool_choice` | `{"mode": "auto" \| "none" \| "required" \| "function", "name": ...}` | | `response_format` | `{"type": "text" \| "json_object" \| "json_schema", ...}` | Output: `response`, `reasoning`, `tool_calls`, `usage`. Each app also has an `openai` function that takes Chat Completions `messages`: `belt app sample google/gemini-4-argon -f openai`. ## Examples ### Deep Analysis with Argon ```bash belt app run google/gemini-4-argon --input '{ "system_prompt": "You are a securities lawyer. Cite the clause for each finding.", "text": "Review this term sheet for investor-unfriendly provisions", "files": ["https://example.com/term-sheet.pdf"], "reasoning_effort": "high" }' ``` ### Agentic Coding with 3.8 Flash ```bash belt app run google/gemini-3-8-flash --input '{ "system_prompt": "You are a Python engineer. Return a unified diff.", "text": "Add retry with exponential backoff to this HTTP client: ", "reasoning_effort": "medium" }' ``` ### High Throughput with Flash-Lite ```bash belt app run google/gemini-3-5-flash-lite --input '{ "text": "Translate to Spanish: The meeting is moved to Thursday.", "reasoning_effort": "none", "max_tokens": 100 }' ``` Run many inputs at once: ```bash # one JSON input per line belt app run google/gemini-3-5-flash-lite --batch inputs.jsonl ``` ### Vision ```bash belt app run google/gemini-3-8-flash --input '{ "text": "Describe this chart and list the three largest values", "images": ["https://example.com/chart.png"] }' ``` ### Multi-turn Conversation ```bash belt app run google/gemini-4-argon --input '{ "context": [ {"role": "user", "text": "I am planning a 5-day trip to Kyoto."}, {"role": "assistant", "text": "Great choice. What are your interests?"} ], "text": "Temples and food. Draft an itinerary." }' ``` ### Structured Output ```bash belt app run google/gemini-3-5-flash-lite --input '{ "text": "Extract product and price: The Pixel case costs $29.", "response_format": { "type": "json_schema", "name": "product", "json_schema": { "type": "object", "properties": {"product": {"type": "string"}, "price": {"type": "number"}}, "required": ["product", "price"] } } }' ``` ### Tool Use ```bash belt app run google/gemini-3-8-flash --input '{ "text": "Find open issues labeled bug", "tools": [{ "type": "function", "function": { "name": "search_issues", "description": "Search the issue tracker", "parameters": {"type": "object", "properties": {"label": {"type": "string"}, "state": {"type": "string"}}, "required": ["label"]} } }], "tool_choice": {"mode": "auto"} }' ``` ## Pricing | Model | Input | Output | |-------|-------|--------| | Gemini 4 Argon | $2.00/M | $10.00/M | | Gemini 3.8 Flash | $1.50/M | $7.50/M | | Gemini 3.5 Flash-Lite | $0.30/M | $2.50/M | Estimate a run: `belt app estimate --input input.json`. ## Related Skills ```bash # Full platform skill (all apps) npx skills add inference-sh/skills@infsh-cli # All LLMs in one place npx skills add inference-sh/skills@llm-models # Other providers npx skills add inference-sh/skills@claude npx skills add inference-sh/skills@gpt # Agents and tool use npx skills add inference-sh/skills@agent-tools # RAG pipelines npx skills add inference-sh/skills@ai-rag-pipeline ``` Browse all apps: `belt app list` ## Documentation - [Agents Overview](https://inference.sh/docs/concepts/agents) - Building AI agents - [Running Apps](https://inference.sh/docs/apps/running) - How to run apps via CLI