Gemini 4 Argon: Google's New Frontier Model for Coding, Enterprise ...

Gemini 4 Argon: Google's New Frontier Model for Coding, Enterprise Work, and Cyber Defense

Google officially unveiled Gemini 4 Argon on September 30 — a model built to sustain deep reasoning through long, complex tasks, with particular strength in software engineering, enterprise work like legal and finance, and cybersecurity defense. Access starts with trusted cyber defenders through Google's Fairwind Program, ahead of a wider rollout to developers, enterprises, and Google AI Ultra subscribers.

What stands out:

The model's output ceiling grows 15x, from 64,000 to 1 million tokens, letting it work through much longer, multi-step problems without stopping. On coding, it sets a new state of the art on DeepSWE v1.1 (77.9%) and is already driving internal projects at Google — including migrating massive C/C++ codebases to Rust (such as Fuchsia's 800,000+-line Zircon kernel) and cutting a quantum-computing optimization benchmark by 40% in minutes. For enterprise work, it tops the Vals Index (which measures economic impact across finance, coding, legal, and tax), leads Harvey's Legal Agent Benchmark and Vals Finance Agent v2, and ranks #1 on Zapier's AutomationBench at 51.3%. It's also state-of-the-art on long-video understanding, scoring 91.7% on LVBench.

On cybersecurity, Argon ties for first place on CWE-bench v1 (68%) for patching vulnerabilities, and security firm Wiz — using it through its free "Scan for Good" initiative — caught a critical flaw in hospital software worldwide that earlier frontier models had missed. Trusted defenders get access without the usual cyber guardrails, to make full use of its offensive and defensive security capabilities.

On safety, Google says it's strengthened defenses against misuse (cyber and CBRN risks), improved resistance to prompt injection (leading Gray Swan's benchmark for it), added chain-of-thought monitoring to catch misalignment, and hardened its training environments — while also going through the U.S. government's voluntary pre-release review process.

Pricing starts at an introductory $2 per million input tokens and $10 per million output tokens, rising to $4/$20 once that period ends. Internally, Google says thousands of employees are already using Argon day to day, including freeing over 300 TiB of data-center memory, with an estimated 500 TiB to 1 PiB in total savings once fully rolled out.