Google unveils Gemini 4 Argon: its new frontier model for coding and cyber defense — defenders get first access

#What happened

Google DeepMind launched Gemini 4 Argon on September 30, 2026 — the first flagship of the Gemini 4 generation. Announced by Google DeepMind SVP and Chief AI Architect Koray Kavukcuoglu, Argon is built for long-horizon, multi-step work in three areas: software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense. The model supports an output window of up to 1 million tokens, up from 64,000 on the prior generation. Introductory pricing is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the intro window, with cached input tokens discounted 95% below the input price. On Google's own published results, Argon scored 77.9% on DeepSWE v1.1 (ahead of Claude Opus 5.5 at 74.2% and GPT-6 Astra at 74.1%), tied for first at 68% on CWE-bench v1 for vulnerability remediation, and hit 51.3% on AutomationBench for end-to-end business tasks — leading or tying on 13 of 18 disclosed benchmarks against GPT-6 Astra and Claude Opus 5.5. This is a gated release: Argon is rolling out first to a set of trusted cyber defenders through Google's Fairwind Program, with wider access for paid API customers and Google AI Ultra subscribers to follow. No broader release date was announced. Google says it is also giving the model to the U.S. government under its voluntary pre-release model access process, and thousands of Google employees already use it for coding, research, and writing.

#Why it matters

Google has just entered the frontier-model slugfest on OpenAI's own terms: Argon's published scores land directly against GPT-6 Astra and Claude Opus 5.5, and the pricing undercuts both at the introductory tier. The defense-first rollout is the real tell — the same week Anthropic warned that China's open-weight GLM-5.3 nearly matches Claude Mythos Preview at building cyber exploits, Google is gating its own strongest model to trusted defenders first. Note the caveat: Bloomberg reported the same day that some Google employees with direct access say Argon performs less well on real coding tasks than the benchmarks suggest; Google told Bloomberg the characterization is inaccurate.