Summary
Google has announced Gemini 4 Argon, a model it says is built for complex, long-running coding and enterprise tasks. Its output limit rises from 64,000 to 1 million tokens, while access is initially being rolled out to trusted cyber defenders.
Google announced Gemini 4 Argon on September 30, 2026, describing it as a model for complex, extended tasks in software engineering, enterprise knowledge work and cybersecurity. The company says the model’s output limit has increased from 64,000 to 1 million tokens. Its initial rollout is to trusted cyber defenders through the Fairwind Program.
Google presents Argon as a model that can sustain multi-step work, rather than only answer a single prompt. The performance figures and examples below are results and uses reported by Google.
A larger output budget for long tasks
A token is a unit of text processed or generated by a language model; it does not correspond exactly to a word. The 1-million-token figure announced for Argon is an output limit: it concerns how much the model can generate, not a stated input-context limit. Google says the expanded budget is intended to let Argon work through difficult problems and produce longer responses or reasoning trajectories in one run.
Google reports results across several evaluations. It says Argon scored 77.9% on DeepSWE v1.1, a benchmark for long-horizon software engineering tasks, and ranked first on Zapier’s AutomationBench with a score of 51.3%. Google also describes Argon as the leading model on the Vals Index, which assesses work in areas including finance, coding, legal services and tax, and reports a score of 91.7% on LVBench, an evaluation of long-video understanding.
The announcement also describes Google’s internal software work. For libgav1, its open-source video decoder, Argon agents replaced 32,000 lines of SIMD code in an existing Rust port with safe Rust. Google says the resulting decoder produced identical video output and ran 2.7 times faster than that Rust port, bringing its performance closer to the optimised C++ version. Google says large migrations, including work on critical codebases, go through automated and manual audits, emulation testing and review before production rollout.
Cybersecurity capabilities and phased access
Google says Argon can autonomously find, validate and patch critical software vulnerabilities. On CWE-bench v1, which evaluates vulnerability remediation, the company reports that Argon tied for first place with a score of 68%.
The announcement describes an early use by Wiz through its Scan for Good initiative, which focuses on finding and remediating high-risk exposures in critical public infrastructure. Google says Argon identified a serious vulnerability in healthcare software used by hospitals worldwide that could expose sensitive personal information, and that previous frontier models had missed it.
For trusted defenders and Google’s own teams, the company says it will release Argon without cyber guardrails so they can use its cybersecurity capabilities. Wider access is being phased in: Google says it is gathering feedback from early testers and iterating on safeguards before making the model available to developers, enterprises and consumers. It also says it is participating in a voluntary US government process for pre-release model access.
Google announced introductory token pricing of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced 95% below the input-token price. The announcement does not give a calendar date for broader availability.