Google on September 30 announced Gemini 4 Argon, the first model in its Gemini 4 generation, describing it as a frontier model for real-world software engineering, enterprise knowledge work and cyber defense. The announcement came in a blog post from Koray Kavukcuoglu, Google's chief AI architect. The model is not yet broadly available.

Access is restricted at launch

Argon is initially being offered to "trusted cyber defenders" through Google's Fairwind Program, the early-access initiative the company launched in early September alongside Gemini 3.8 Flash Cyber. Google says it is also taking part in the U.S. government's voluntary pre-release model access process, and that broader availability will follow, starting with paid API customers and Google AI Ultra subscribers. It gave no date. According to TechCrunch, Google staff are already using the model internally for debugging and codebase migration.

Google framed the staged rollout around safeguards in four areas: misuse defense, prompt-injection robustness, misalignment monitoring and hardened sandboxed environments.

What Google claims

Google says Argon scores significantly higher than OpenAI's GPT-6 Astra and Anthropic's Fable and Opus models across a range of benchmarks. Figures published by Google include 77.9% on DeepSWE v1.1, 51.3% on AutomationBench, 91.7% on LVBench for long-video understanding, and a tie for first at 68% on CWE-bench v1, a vulnerability-remediation test. It also cites leadership on the Vals Index, which covers finance, coding, legal and tax work. The model supports up to 1 million output tokens, up from 64,000 on previous Gemini models.

VentureBeat's tally of Google's disclosures found Argon leading or tied on 13 of 18 benchmarks, with examples including 77.9% on DeepSWE v1.1 versus 74.2% for Claude Opus 5.5, and 51.3% on AutomationBench versus 42.5%. GPT-6 Astra still leads on some tests, notably FrontierSWE v2 and Terminal-Bench Science. These are vendor-reported results; independent testing is limited while access is restricted.

Pricing

Google lists introductory API pricing of $2 per million input tokens and $10 per million output tokens, moving to $4 and $20 afterward, with a 95% discount on cached input tokens. VentureBeat notes that undercuts GPT-6 Astra's $10 and $50 and matches Claude Opus 5.5 once the introductory period ends. Because most developers cannot yet call the model, whether the benchmark gains and pricing hold up on production workloads remains untested.

Why it matters

Argon follows a pattern set by OpenAI's Astra and Anthropic's Mythos models: the most capable cyber-relevant systems reach vetted defenders first, with general release gated behind safeguards and government review. For Gemini users, the practical change is still ahead; the headline is that Google is again competing for the top of the leaderboard.