Google has introduced Gemini 4 Argon but is not making the frontier model broadly available yet. Initial access is going to selected cybersecurity defenders through the Fairwind Program while Google gathers feedback, tests safeguards and participates in a US government pre-release review process.
Argon is designed for long-running coding, legal, financial and security work. Its maximum output expands from the previous 64,000 tokens to one million, allowing a single task to continue through much longer reasoning and generation. Google claims a 77.9 percent score on the DeepSWE v1.1 software-engineering benchmark and says internal agents have handled large code migrations, though those results remain company-reported.
Cybersecurity is the most sensitive part of the launch. Trusted defenders will receive a version without normal cyber guardrails so they can find and patch vulnerabilities. Google says broader access depends on protections against misuse, prompt injection and agents acting outside their assigned scope.
When Argon reaches paid API customers and Google AI Ultra subscribers, introductory pricing is planned at $2 per million input tokens and $10 per million output tokens, with cached inputs discounted by 95 percent. Google has not given a firm date for that wider rollout.