Gemini 4 Argon launches: 1 million tokens will be distributed, with the first batch reserved exclusively for trusted security teams.

Beating AI Express: Google DeepMind has launched its next-generation flagship model, Gemini 4 Argon. Designed for long-horizon complex tasks, it covers software engineering, finance, legal work, and cybersecurity. The model is not yet available to general users; initially, it will only be provided to trusted cybersecurity teams via the Fairwind Program, with plans to gradually roll out access to paid API clients and Google AI Ultra users later. One of Argon’s key upgrades is its output length: the single-turn output limit has been raised from the previous 64K tokens to 1 million tokens, enabling the model to sustain reasoning and execute longer steps in a single task. Google says it has already used the model internally for large-scale code migration, algorithm optimization, and data center memory optimization. In official evaluations, Argon scored 77.9% on DeepSWE v1.1, outperforming GPT-6 Astra (74.1%) and Claude Opus 5.5 (74.2%). It also ranked first on the Vals Index and AutomationBench. However, it is not fully leading: on FrontierSWE, Terminal-Bench 4.0, and some scientific tasks, it still lags behind GPT-6 Astra or Claude Opus 5.5. Google has also significantly strengthened its cybersecurity capabilities: Argon can automatically discover, verify, and patch software vulnerabilities. The version provided to trusted defense teams even removes some cybersecurity restrictions to unlock its full functionality. Ahead of full launch, Google is continuing to test risks including abuse, prompt injection, and agent overreach. During the initial API rollout, pricing will be $2 per million input tokens and $10 per million output tokens, with prices later doubling to $4 and $20 respectively.