Google announces Gemini 4 Argon for coding, knowledge work and cyber defense

Google announced Gemini 4 Argon on September 30, 2026, describing it as a new frontier model for software engineering, professional knowledge work and cybersecurity defense. The announcement marks the next major generation in Google’s Gemini model family. Argon is not broadly available at launch: Google says it is first rolling out to a group of trusted cyber defenders through its Fairwind Program, while the company continues to refine safeguards. Wider access is planned for developers, businesses and consumers, starting with paid API customers and Google AI Ultra subscribers. Google has not given a public release date for those groups.

A notable change is the model’s output limit. Google says Gemini 4 Argon can generate up to one million output tokens in a single run, compared with 64,000 previously. That is an output limit, not a promise that every answer will use that much text or that the model can process a million input tokens. The longer generation window is intended for tasks that may involve sustained reasoning, such as large code changes or extended research. Google also lists introductory API prices of $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted by 95 percent. Those prices are announced ahead of broad API availability.

Google presents examples from its own engineering work: Argon agents helped migrate C and C++ code to Rust, and the company says a rewritten Rust video decoder ran 2.7 times faster than an existing Rust port. Google also reports that the model led on DeepSWE v1.1, a benchmark for long-running software engineering tasks, and tied for first on CWE-bench v1 for fixing vulnerabilities. These are company-reported results; they do not establish how the model will perform across independent workloads.

The cybersecurity rollout is especially cautious. Google says selected defenders will get access without its cyber guardrails so they can search for and repair vulnerabilities, while the company strengthens protections before wider release. That could help security teams find weaknesses sooner, but the same capabilities require careful access controls. For developers and businesses, the practical question is whether longer autonomous work and the announced price translate into reliable results on their own tasks. Until broader access and independent evaluations arrive, Argon is a substantial announcement with limited hands-on evidence outside Google and its initial testers.