Google’s reasoning model for complex coding, source-grounded analysis, and visual understanding.
Use google/gemini-4-argon through the sjolt LLM API.
google/gemini-4-argonToken rates
google/gemini-4-argonSimilar Models
An Anthropic model for careful coding, long-document analysis, and visual understanding with optional adaptive thinking.
An Anthropic model for coding, long-context knowledge work, and visual understanding with adaptive thinking.
An Anthropic model for responsive coding, long-document analysis, and text-and-image understanding with adaptive thinking.
Model overview
Review substantial implementations, connect source material, and interpret images with adjustable reasoning through OpenAI-compatible Chat Completions.
Provide the relevant code, expected behavior, constraints, and failing cases. Ask for a focused diagnosis, a proposed change, and checks that would distinguish a correct solution from a plausible one.
The 1,048,576-token context window is shared by input and output. Label pasted documents and conversation history, reserve space for the answer, and ask the model to distinguish evidence from assumptions.
Send base64 image data URLs through the API with a specific question or comparison task. Ask for observable details before conclusions. Responses are text; the artwork on this page is conceptual and is not output from Gemini 4 Argon.
Model characteristics
Input and output share 1,048,576 tokens. This Chat Completions integration accepts up to 65,536 output tokens, including thinking.
Reasoning remains enabled. Set reasoning_effort to low, medium, high; the default is medium. Use clear requirements and explicit completion criteria for longer tasks.
Send model and messages to /v1/chat/completions. Keep system instructions, assistant tool calls, and tool results in their native message format when continuing a conversation.
FAQ
In native usage, completion_tokens counts non-thinking output and completion_tokens_details.reasoning_tokens separately counts thinking. Add these values for the complete output count.
Complex software engineering, mathematical and scientific reasoning, multi-step tool workflows, and analysis grounded in supplied source material.
Send text and image_url content blocks with base64 data URLs through Chat Completions. The Playground accepts text. The response is text.
Use reasoning_effort with low, medium, high. The Playground starts at medium; reasoning stays enabled at every level.
The model has a unified 1,048,576-token window for input and output. This integration accepts an output limit up to 65,536 tokens. Leave room for both thinking and the final answer.
Use model google/gemini-4-argon through OpenAI-compatible Chat Completions. API requests may stream with stream: true; the Playground uses non-streaming responses.