Overview
- Google announced Gemini 4 Argon on September 30 and has begun a phased rollout that gives early access to trusted cyber defenders through its Fairwind Program and to U.S. pre-release reviewers.
- Argon raises the model output limit to 1 million tokens so it can sustain very long, multi-step tasks in a single run and, according to Google, leads many internal benchmarks versus rival frontier models.
- Google says Argon was trained for defensive cybersecurity and that partner Wiz used it to find a critical vulnerability in healthcare software; Fairwind participants get the model without some cyber guardrails so they can use its full vulnerability‑hunting ability.
- Introductory API pricing is $2 per million input tokens and $10 per million output tokens with planned increases later to $4 and $20, and Google says paid API customers and AI Ultra subscribers will be next in line for access though no general release date has been set.
- Thousands of Google employees have already used Argon internally to free hundreds of tebibytes of data‑center memory and speed algorithm work, and the company says it is iterating on guardrails, red‑teaming and sandboxing before broader availability.