Gemini 4 Argon Reaches General Availability in the Gemini API
Gemini CLIGoogle released Gemini 4 Argon (gemini-4-argon) as a generally available model in the Gemini API, positioning it as the top-tier option for demanding coding, agent-driven, math and science workloads. The model offers a 1,048,576-token context window, configurable thinking levels, and support for function calling, code execution and Google Search grounding. Developers can use it in Google AI Studio and through the Gemini API.
Key Takeaways
- Gemini 4 Argon is now generally available in the Gemini API as
gemini-4-argon, and Google positions it as its top-tier model for demanding workloads. - The model targets coding, autonomous agents, math and science, making it the natural choice for complex software engineering tasks.
- A 1,048,576-token context window is shared between input and output, with long decode allowing up to 1,048,576 output tokens.
- Low, medium and high thinking levels let developers tune reasoning depth against latency and cost.
- Built-in tools include function calling, code execution, file search, URL context and Google Search or Maps grounding, with computer use in preview.
- It runs on Batch, Flex and Priority inference, but has no Live API, image generation or explicit caching support.
Gemini 4 Argon Is Now Generally Available
Google announced on October 13, 2026 that Gemini 4 Argon has reached general availability in the Gemini API under the model ID gemini-4-argon. The Gemini API release notes describe it as the top-tier option for demanding coding, agent-driven, math and science workloads, and the model page calls it a flagship reasoning model built for complex software engineering, advanced math and science, autonomous multi-step agents and long-decode tasks. The model is usable in Google AI Studio, and Google points developers to a dedicated "What's new in Gemini 4 Argon" page for API changes, pricing and migration guidance.
Context Window and Output Limits
Gemini 4 Argon provides a context window of 1,048,576 tokens, shared between input and output. Synchronous requests produce up to 65,536 output tokens by default, while the long decode option raises the ceiling to 1,048,576 output tokens. That makes the model suited to very long generations such as large code changes, extended reasoning traces and multi-step agent runs.
Inputs, Outputs and Thinking
The model accepts text, image, video, audio and PDF input and produces text output. Thinking is supported at low, medium and high levels, letting developers trade latency and cost against reasoning depth. Audio generation, image generation and the Live API are not supported.
Tools and Agent Features
Gemini 4 Argon supports function calling, code execution, file search, structured outputs, URL context, Google Search grounding and Google Maps grounding. Agentic video understanding is also supported and defaults to a static mode. Computer use is available in preview. Implicit caching is supported, while explicit caching is not.
Consumption Options
The model works with the Batch API, Flex inference and Priority inference, so teams can choose between lower-cost asynchronous processing and faster, higher-priority serving depending on the workload.
What Developers Should Know
Developers moving from earlier Gemini models should review the migration guidance linked from the Gemini API documentation before switching model IDs. Pricing, regional availability and any benchmark figures are published on Google's pricing and model pages rather than in the release note itself, so those pages are the place to confirm costs before adopting the model in production.