Gemini 4 Argon Review

Gemini 4 Argon is the model Google announced on September 30, 2026 as the start of its next frontier generation, and the first thing to know is that you almost certainly cannot use it. It is rolling out only to trusted cyber defenders through Google's Fairwind Program, and to them without cyber guardrails, while Google takes part in the U.S. government's voluntary pre-release testing and keeps hardening its safeguards. Paid API customers and Google AI Ultra subscribers are promised access first when it widens. There is no date.

The claims are strong. On Google's figures Argon sets a new high on DeepSWE v1.1, 77.9 percent on long-horizon software engineering, leads the Vals Index across finance, coding, legal and tax work, ranks first on Zapier's AutomationBench at 51.3 percent and ties for first on CWE-bench v1 for patching vulnerabilities. Google's own comparison is not a clean sweep: Claude Opus 5.5 beats it on Terminal-Bench 4.0, 66.4 percent to 57.4, and GPT-6 Astra beats it on FrontierSWE.

Inside Google, Argon agents are moving C and C++ libraries to Rust, including a libgav1 video decoder that now runs 2.7 times faster than the earlier Rust port, and the output limit rises to a million tokens from 64 thousand, so one run can think and write at length. The launch price is aggressive for a frontier model, $2 in and $10 out per million tokens with cached input 95 percent off, rising to $4 and $20 after an introductory period.

None of it can be checked from outside yet. Every number above is Google's or a partner's, and the model is not in the public API, the Gemini app or anything a reader can open, which is why ease of use sits at a C and is held there by name. On what Google has shown, Argon belongs with the best models on this list. Until people can use it, an A is the honest grade, and it will be re-reviewed when access opens.