Google launched Gemini 4 Argon, a model for coding, research and writing, trained specifically for defensive cyber work: Google says it can “autonomously find, validate, and patch critical software vulnerabilities.” It is rolling out first only to select cyber partners through Google's Fairwind Program.
Google says its own staff already use Argon daily for debugging and codebase migrations, and cites Vals benchmarks showing it scoring significantly higher than GPT-6 Astra and Anthropic's Fable and Opus models. Those claims are Google's own and not independently verified.
Google's Gemini 4 Argon announcement — Primary source for benchmark claims.
llms.txt: agent-optimization wishful thinking — Skeptical view relevant to GEO advice.
Holo4 computer-use agents — Open computer-use agent release.