Google DeepMind has announced Gemini 4 Argon, the first model in its Gemini 4 generation, designed for long-horizon software engineering, enterprise knowledge work, and cybersecurity defense. The primary technical advancement is a maximum output length of 1 million tokens per response, significantly exceeding the 128K limits of current frontier models like Claude Opus 5.5 and GPT-6 Astra.
- Argon leads on DeepSWE v1.1 (77.9%) and Vals Index (68.9%), while trailing GPT-6 Astra on FrontierSWE v2 and Terminal-Bench 4.0.
- Introductory pricing is set at $2 per 1M input tokens and $10 per 1M output tokens, with a 95% discount for cached inputs.
- The model is currently available through the Fairwind program for trusted cyber defenders and early testers before a wider release.
The increased output capacity allows developers to generate large refactors or long reports in a single trajectory without splitting work across multiple turns.