Google releases Gemini 4 Argon to trusted cyber defenders only

Google announced its new frontier model Gemini 4 Argon on 30 September 2026, releasing it first to a set of trusted cyber defenders through the Fairwind Program. The model targets real-world software engineering, enterprise knowledge work in legal and finance, and cybersecurity defense. Koray Kavukcuoglu, Senior Vice President at Google DeepMind and Chief AI Architect, described the release as a phased approach to safely shipping frontier capabilities.theverge.com+1

Google is holding back broader access deliberately. The company said it is actively engaged in the U.S. government's voluntary process for pre-release model access while it gradually expands who can use the model. Google plans to gather feedback from early testers and iterate on guardrails before making Argon available to developers, enterprises, and consumers. The company said it will strengthen critical frontier safeguards, including defending against misuse and prompt injection attacks and monitoring for misalignment.thenewstack.io

Why access is restricted at first

Argon extends the model's output token limit to 1 million tokens, up from the previous 64,000 tokens. A token is a unit of text the model reads or writes. Google said the extra headroom lets the model think deeply and generate hundreds of thousands of tokens in a single trajectory, adding depth to reasoning on hard problems. The company also said thousands of Googlers already use Argon internally for specialized coding, deeper research, and writing quality.

Google reported concrete internal results. Argon helped quantum computing researchers optimize the spacetime resources of subroutines, beating a published baseline by 40% in minutes. A team of Argon agents identified and applied memory optimizations across Google's data centers, freeing over 300 TiB of memory once rolled out, with an estimated 500 TiB to 1 PiB in total savings. Argon agents are also migrating C/C++ codebases to Rust, scaling up to over 800,000 lines for the Fuchsia OS Zircon kernel.

For libgav1, Google's open source video decoding software, Argon agents replaced 32,000 lines of SIMD code by running many rounds of profile-guided experiments and producing safe Rust. The result runs 2.7 times faster than the Rust port with identical video output. On DeepSWE v1.1, which measures performance on real-world long-horizon software engineering tasks, Argon sets a new state of the art at 77.9%.

What the model can do

Beyond coding, Google said Argon leads on the Vals Index, which measures economic impact across finance, coding, legal, and tax work with each sector weighted by its contribution to U.S. GDP. Argon also ranks first on Zapier's AutomationBench, which measures end-to-end execution across core business functions, with a score of 51.3%. Google pointed to strong results on Vals Finance Agent v2 and Harvey's Legal Agent Benchmark.

The New Stack assessed the release as strong on knowledge work benchmarks but mixed on coding scores, noting that access is limited. The publication's framing matches Google's own description of a restricted first phase. Google shared a chart showing many benchmarks where Gemini 4 fares better than competing models from OpenAI and Anthropic, and noted that the AI community had been discussing what appeared to be leaked benchmarks on X earlier that week.

Google set an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input token price. The launch came a day after OpenAI's DevDay conference, where OpenAI launched a Dots AI agent and the GPT-6.1 Sol AI model. OpenAI also said that week it would not release a planned GPT-6.1 Astra model over safety worries.

Where the benchmark numbers came from

Kavukcuoglu took over as head of DeepMind in August, making this one of his first major model announcements in the role. Google said Argon is already powering internal workflows and that the company is iterating on guardrails with early tester feedback. The company framed the phased release as necessary for shipping frontier capabilities safely rather than as a permanent limitation on who can use the model.

Google has not announced a specific date for wider availability. The company said it will make Argon available to developers, enterprises, and consumers as soon as possible, after gathering early tester feedback and iterating on guardrails. The next step is expansion of the Fairwind Program beyond the initial set of trusted cyber defenders, with the U.S. government's voluntary pre-release access process continuing in parallel.