Google DeepMind has unveiled Gemini 4 Argon, its newest frontier model, with an output limit expanded sixteen-fold to one million tokens from the previous 64,000. The company says this is intended to support long-horizon reasoning, letting the model work through much longer chains of analysis or generated code without being cut off mid-task.
Rather than a broad public launch, Argon is rolling out first to a limited group of trusted cyber defenders through Google's Fairwind Program, alongside internal Google teams. For that group, Google says it is releasing the model without its usual cyber guardrails, reasoning that defenders need the full capability to find and close vulnerabilities before attackers exploit the same gaps. Security firm Wiz is already using Argon through its Scan for Good initiative. On CWE-bench v1, a benchmark for vulnerability remediation, Google reports a score of 68%, which it says ties for the top result.
For coding, Google cites a 77.9% score on DeepSWE v1.1, though Bloomberg has reported that some Google employees say Argon performs less well on real-world coding tasks than the benchmark number implies; Google disputes that characterisation.
Paid API access and Google AI Ultra subscribers are expected to follow the cyber-defender rollout, with Google saying it has not yet set a date for a wider public release. Introductory pricing is reported at $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached input tokens.
Source: MarkTechPost
