Gemini 4 Argon arrives cheaply—but only through a controlled cyber rollout

Google has announced Gemini 4 Argon, its first new flagship frontier model in nearly a year. The company says it is strong in coding, cyber defence and long-running legal and financial workflows. Those are vendor claims for now: the public cannot independently test the model, and Google has not given a date for general access.

Argon is initially available to selected cybersecurity specialists through Google’s Fairwind programme, where they can use it to find and fix software vulnerabilities. Google says it is also participating in the US government’s voluntary pre-release access process and will widen availability only after feedback and additional safeguards. The Information describes the release as similar to other labs’ phased frontier rollouts.

The introductory API price is unusually aggressive: $2 per million input tokens and $10 per million output tokens, with cached inputs discounted by 95%. Google says the rates will later double to $4 and $20—still below the top models from OpenAI and Anthropic cited by The Information. Paid API customers and Google AI Ultra subscribers are due to receive access first when the wider release begins.

The sequencing matters as much as the price. Cyber capability is intrinsically dual-use: the same reasoning that identifies a vulnerability can help exploit it. Restricting the first deployment to identifiable experts creates a controlled environment for measuring misuse and failure before mass distribution. But the safety case will depend on evidence Google has not yet published—what evaluations it runs, what access can be revoked and whether the safeguards hold outside Fairwind.