Google announced Gemini 4 Argon on 30 September, a frontier model it describes as built for “real-world coding, enterprise knowledge work, and cyber defense”. It is not going to developers first. Initial access goes to a set of trusted cyber defenders through what Google calls its Fairwind Program, and the company says it is “actively engaged in the U.S. government’s voluntary process for pre-release model access” while it widens availability. Paid API customers and Google AI Ultra subscribers are next in line.
The benchmark claims are Google’s own and none has been reproduced independently: 77.9% on DeepSWE v1.1, 68% on CWE-bench v1 (a tie for first place), 51.3% on Zapier’s AutomationBench and 91.7% on LVBench for long video understanding. Google also says Argon leads the Vals Index and Gray Swan’s indirect prompt injection benchmark.
The change developers will notice most is the output limit, which Google is raising to 1M tokens from 64K. Introductory pricing is $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20.
Staging a frontier release through security researchers and a government pre-release process, rather than a general launch, is a shift in how the largest labs expect to ship. The cyber defence claim depends on evaluations nobody outside the Fairwind group can run yet.