Google announced Gemini 4 Argon on Sept. 30, describing it as a frontier model for long-horizon software engineering, enterprise knowledge work and cybersecurity defense. Initial access is limited to trusted cyber defenders in Google’s Fairwind Program while the company participates in the U.S. government’s voluntary pre-release model-access process, according to Google’s DeepMind blog and Ars Technica (opens in new tab).

Google said Argon scores 77.9% on DeepSWE v1.1, ahead of Anthropic’s Claude Opus 5.5 (74.2%) and OpenAI’s GPT-6 Astra (a tenth of a point lower) on Google’s own chart, SiliconANGLE (opens in new tab) reported, and leads the Vals Index for economic-impact tasks across finance, legal and related work. Output context expands to 1 million tokens from 64,000 previously. Introductory API pricing is listed at $2 per million input tokens and $10 per million output tokens (cached inputs 95% off), rising to $4/$20 after the intro period.

Internally, Google said Argon agents have freed more than 300 TiB of data-center memory and are migrating C/C++ codebases to Rust, including more than 800,000 lines in Fuchsia’s Zircon kernel; those figures are company-reported. Google also said Argon, used by its Wiz subsidiary in Wiz’s free Scan for Good program, uncovered a critical vulnerability exposing personal information in healthcare software used by hospitals worldwide that earlier frontier models had missed; Google did not provide specifics, Ars Technica (opens in new tab) noted.

Google said Argon monitors chain-of-thought for misalignment and strengthens refusals against cyber and CBRN misuse, while releasing the model to Fairwind partners without its cyber guardrails so defenders can use its full capability. Broader availability is planned for paid API customers and Google AI Ultra subscribers after more testing; Google gave no date.