Google opens Gemini 4 Argon to cyber defenders only

Google announced Gemini 4 Argon on 30 September and is releasing it in phases, starting with vetted cybersecurity defenders through its Fairwind Program rather than with developers or consumers.

"Safely releasing frontier capabilities at this level requires a phased approach," wrote Koray Kavukcuoglu, senior vice president at Google DeepMind and the company's chief AI architect. Google says it is taking part in the United States government's voluntary process for pre-release model access, and will widen access "as soon as possible" after further work on guardrails.

The Fairwind Program opened on 3 September with the smaller Gemini 3.8 Flash Cyber model and has signed up more than 650 organisations, including CrowdStrike and Palo Alto Networks, according to SiliconANGLE. Fairwind members and Google's internal teams get a version with the cyber guardrails removed, which can find and fix software vulnerabilities.

Argon is built for long-horizon work, and Google set its output limit at 1 million tokens to match, up from 64,000 on earlier Gemini models. It will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input at 95 percent off the input price.

The benchmark numbers are Google's, published in its own charts. Argon scored 77.9 percent on DeepSWE v1.1, which measures long-horizon software engineering, and 51.3 percent on Zapier's AutomationBench against 42.5 percent for Anthropic's Claude Opus 5.5. It ties for first on CWE-bench v1 at 68 percent. SiliconANGLE notes that Terminal-Bench 4.0 still belongs to Opus 5.5, and that of the 18 benchmarks in Google's charts, VentureBeat counted 12 that Argon led outright. Google also says Argon is state of the art on LVBench, a long video understanding benchmark, with a score of 91.7 percent.

The result most relevant to video work is buried in the internal ones. Agents running Argon worked on Google's open-source libgav1 video decoder, taking a Rust port and replacing 32,000 lines of SIMD code with safe Rust the compiler could optimize itself. Google says the decoder now runs 2.7 times faster than the earlier port with no change to its video output. A separate fleet-wide sweep of profiling data by Argon agents freed more than 300 tebibytes of memory across Google's data centres.

For creators the practical read is indirect: none of this is a video generation model, and Argon is not something an individual can call today. What it signals is that the largest labs are now shipping frontier models to narrow, vetted groups first, and that the price of a million-token context is falling fast.

Developers, enterprises and consumers have no access date. Google says it is still iterating on guardrails.