Gemini 4 Argon
A flagship frontier large language model released by Google DeepMind, supporting a maximum output of one million tokens. It is designed for software engineering, legal and financial enterprise knowledge work, and cybersecurity defense scenarios, with capabilities for long-horizon deep reasoning.
Core Facts
Gemini 4 Argon adopts a phased rollout strategy, currently available only through the Fairwind program for cybersecurity defenders and not yet fully open to the public.
Gemini 4 Argon is positioned by Google as a 'workhorse' model built specifically for coding and cybersecurity work.
Google stated that after the introductory period ends, the price of Gemini 4 Argon will rise to $4 per million input tokens and $20 per million output tokens, positioning it at the Opus 5.5 level.
Timeline (last 90 days)
Google does not provide free Argon API access, and any website promising free Argon API access should be regarded as unverified.
Cursor added Gemini 4 Argon on October 13, with pricing consistent with Google's current rates.
Gemini 4 Argon officially opened on October 13.
Gemini 4 Argon only supports implicit caching, not explicit caching; cached input is $0.10 per million tokens before November 30, and $0.20 afterward.
Google provides a free Search grounding quota for Gemini 4 Argon: 5,000 free requests per month (shared by Gemini 3 and subsequent models), and $14 per 1,000 requests afterward.
OpenCode added Gemini 4 Argon to OpenCode Go on October 14, launching a promotion of '$60 of usage for $10 before November 30'.
Gemini 4 Argon's thinking tokens are billed as output, and the key parameter for controlling cost is thinking_level (low, medium, high); higher levels consume more, so use low for simple tasks and only worth setting to high for complex tasks.
The configuration of Gemini 4 Argon leaked in developer tool code shows that the context window has two tiers, 512K and 900K, with quota multipliers of 1.3x and 1.8x respectively.
Many hands-on tests show that Gemini 4 Argon frequently makes errors when performing real-world tasks, and hallucinations still exist in its answers.
In the demo, Gemini 4 Argon was able to generate in one go a 3D interactive webpage containing a day-night cycle, rainfall, volcanic eruptions, and dinosaur herd behavior.
40 more timeline events
All Facts (20)
Gemini 4 Argon adopts a phased rollout strategy, currently available only through the Fairwind program for cybersecurity defenders and not yet fully open to the public.
90%VerifiedGemini 4 Argon is positioned by Google as a 'workhorse' model built specifically for coding and cybersecurity work.
90%VerifiedGoogle stated that after the introductory period ends, the price of Gemini 4 Argon will rise to $4 per million input tokens and $20 per million output tokens, positioning it at the Opus 5.5 level.
80%VerifiedGemini 4 Argon's maximum output has been increased from 64,000 tokens to 1 million (referring to the output limit, not the context window)
80%VerifiedThere is currently no complete official benchmark data from Google to corroborate the claim that Gemini 4 Argon is comparable to the top configuration of Claude.
75%VerifiedGoogle first handed Argon to trusted security experts for early validation through a project called Fairwind
75%VerifiedGemini 4 Argon achieved a new record of 77.9% on the Deep7e long-workflow software engineering benchmark, surpassing Claude Opus 5.5.
75%VerifiedGemini 4 Argon focuses on three high-value scenarios: software engineering, legal and financial enterprise knowledge work, and cybersecurity defense.
75%VerifiedArgon agents are migrating Google's internal C/C++ codebase to Rust, scaling to over 800,000 lines of the Fuchsia OS Zircon kernel.
75%VerifiedSecurity firm Wiz has used Argon through the Scan for Good program, and in an early demonstration the model discovered a critical vulnerability in a medical software used by hospitals worldwide that was leaking sensitive personal information.
75%VerifiedThe tiered model strategy essentially allocates inference compute dynamically according to task complexity; lightweight models can achieve inference speeds 5–10 times faster than flagship models at one-tenth the cost.
75%VerifiedGemini 4 Argon scores 15% on the hallucination benchmark (lower is better), lower than any current frontier model.
70%VerifiedGemini 4 Argon scored 91% on LV Bench (video/long-content understanding), setting a new record.
70%VerifiedGemini 4 Argon's strengths lie in multimodal understanding and everyday foundational tasks, rather than coding or web development.
70%VerifiedOn the VALS Knowledge Task Index, Gemini 4 Argon scored 68.9%, reaching the top of SOTA and surpassing Opus 5.5's 67%.
70%VerifiedOn Automation Bench, Gemini 4 Argon scored 51.3%, surpassing Fable 5.1's 31.4% and Opus 5.5's 42.5%.
70%VerifiedIn a chart published by Google, Argon's Deep SWE score is close to 78%, while Claude Opus 5.5 and GPT-6 Astra are only slightly above 74%.
65%VerifiedArgon's launch price is $2 per million input tokens and $10 per million output tokens, on par with Sonnet 5.5 and just one-fifth the price of GPT-6 Astra; after the promotional period ends, prices will double to $4 input and $20 output.
65%VerifiedGemini 4 Argon claims to support an output capacity of 1 million tokens.
65%VerifiedAccess to Gemini 4 Argon is limited to government users and trusted network defenders within the Fairwind project.
65%