Gemini 4 Argon

What is Gemini 4 Argon?

Gemini 4 Argon is Google's flagship reasoning model, built for complex software engineering, math and science problems and multi-step agents. Use it through the Gemini API for a large refactor or a long agent task with up to 1M tokens of context. It is paid only in the API, and introductory prices double on December 1, 2026.

What can you do with Gemini 4 Argon?

Google positions Argon for deep work on long tasks: large codebase refactoring and optimization, multi-step agent workflows, and finance, legal and science problems. Its examples include helping the Chrome team split a legacy folder of about 6 million lines out of Chromium and an overnight agent run that sped up session preparation in Google's Agent Development Kit by 40%. These are Google's own reports, not independent tests.

In the API it accepts text, images, video, audio and PDFs and returns text. It supports function calling, structured outputs, code execution, search and Maps grounding, URL context, file search and computer use in preview, with low, medium and high thinking levels. It does not generate images or audio and is not available in the Live API.

Try it on a small task

Suggested workflow: open Google AI Studio with a paid API key and give Argon one bounded engineering task, such as explaining how a module works and proposing a refactor plan for it, together with the relevant files. Start with the medium thinking level and compare the plan with what you know about the code.

Check whether it named the right files and risks, then ask for the change as a diff and run your tests. Note the input and output tokens of the request: thinking tokens count as output, so long reasoning runs can cost more than the prompt size suggests.

Google's benchmark results

Google reports that Argon ranks first on the Vals Index (68.9%) and scores 77.9% on DeepSWE v1.1 and 68% on CWE-bench v1. Its own table also shows where it does not lead: GPT-6 Astra scores higher on FrontierSWE v2, Terminal-Bench Science and OSWorld 2.0, and Claude Opus 5.5 scores higher on Terminal-bench 4.0 and PostTrainBench. Several Argon results are self-computed, and results for other models come mostly from their providers.

View image

Gemini 4 Argon pricing and access

Argon has been available since October 13, 2026 through the Gemini API in Google AI Studio and the Gemini platform, for Google AI Ultra subscribers in Google Antigravity, and for Google AI Pro and Ultra subscribers in the Gemini app and AI Mode in Search; subscription prices vary by country. It is not on the API free tier.

Standard API pricing is $2 per million input tokens and $10 per million output tokens, including thinking tokens, through November 30, 2026, and $4 and $20 from December 1. Cached input costs $0.10 per million tokens, rising to $0.20. Batch and Flex cost half the standard rate and Priority costs more. Estimate a typical task at the December prices before building on it.