Google on September 30, 2026 announced Gemini 4 Argon, its new frontier model, in a primary DeepMind blog post by Koray Kavukcuoglu, SVP of Google DeepMind and Google’s chief AI architect. The company said Argon is rolling out first to a set of trusted cyber defenders through Google’s Fairwind Program, with internal Google teams already using the model for specialized coding, research, and long-horizon workflows. Google framed the limited launch as a phased release while it participates in the U.S. government’s voluntary process for pre-release model access and gathers feedback before broader availability.
According to Google, Argon is built for deep reasoning across complex, long-horizon work in software engineering, enterprise knowledge tasks such as legal and finance, and cybersecurity defense. The company said it is expanding the model’s output token limit to 1 million tokens, up from a previous 64,000. Google reported company benchmark claims including a DeepSWE v1.1 score of 77.9%, a leading position on the Vals Index for economic-impact workflows, an AutomationBench score of 51.3%, and 91.7% on LVBench for long video understanding. For defensive cybersecurity, Google said Argon can autonomously find, validate, and patch critical software vulnerabilities, and that trusted Fairwind defenders and Google internal teams will receive Argon without cyber guardrails so they can use its full defensive capabilities. On CWE-bench v1, Google reported Argon tying for first at 68%.
Google also described internal productivity examples it attributes to Argon agents, including quantum-algorithm spacetime optimizations that beat a published baseline by 40% in minutes, fleet memory optimizations that freed more than 300 TiB once rolled out, and large-scale C/C++ to Rust migration work spanning libraries such as re2 and libgav1 up to hundreds of thousands of lines in Fuchsia’s Zircon kernel. Introductory API pricing was listed at $2 per million input tokens and $10 per million output tokens, with cached input at a 95% discount; after the introductory period, Google said pricing moves to $4 and $20 per million input and output tokens. Before a broader rollout, Google said it is strengthening safeguards against misuse and CBRN risks, improving prompt-injection robustness, monitoring for misalignment via chain-of-thought and action checks, and hardening sandboxed evaluation environments.
DigiEditorial verified the announcement against Google’s September 30 Introducing Gemini 4 Argon post as the primary source, with corroboration from The Verge and SecurityWeek. Benchmark scores, internal productivity figures, Fairwind access details, and pricing are Google’s disclosed claims and secondary summaries of those claims, not independent audits. Broader access for developers, enterprises, and consumers is promised to follow the Fairwind and trusted-tester phase, starting with paid API customers and Google AI Ultra subscribers.