Overview
- Google unveiled Gemini 4 Argon on Sept. 30 and began a limited rollout to vetted cybersecurity partners through its Fairwind program while keeping broader access closed for more testing.
- The company published self‑reported benchmark claims including a DeepSWE v1.1 score of 77.9% that leads several tests while also conceding the model trails on some coding benchmarks.
- Argon raises practical limits for long tasks by supporting up to 1 million output tokens, allowing the model to generate hundreds of thousands of words in a single reasoning trajectory.
- Google set introductory API pricing at $2 per million input tokens and $10 per million output tokens with rates rising to $4 and $20 after the discount period, and it will prioritize paid API customers and AI Ultra subscribers for early access.
- Google says Argon is already used internally for tasks such as freeing more than 300 TiB of data‑center memory and large code migrations, while the company is strengthening safeguards like prompt‑injection defenses, sandbox hardening, and monitoring for misuse before wider release.