Google DeepMind announces Gemini 4 Argon, rolling out first to trusted cyber defenders via Fairwind Program, with broader release pending

Google DeepMind announced Gemini 4 Argon, a new frontier model, but released it only in limited form to trusted cyber defenders through the Fairwind Program and to internal Google teams. Those users get it without the usual cyber guardrails. Google says the output limit rises from 64K to 1M tokens and plans introductory pricing of $2 per million input tokens and $10 per million output tokens, with cached input 95% off. Google reports 77.9% on DeepSWE v1.1 (claimed state of the art), 51.3% on AutomationBench, 91.7% on LVBench and 68% on CWE-bench v1 (tied first). There is no date for broad availability, and the pricing is described as planned, not live.

For most developers and enterprises nothing is usable yet, so there is nothing to buy, test or migrate. The pricing preview and the 1M output limit let teams start modelling costs for long agentic coding and document tasks, but only if the model ships as described. The more consequential detail is the staged release of a model with relaxed cyber guardrails to vetted defenders, alongside participation in the US government's voluntary pre-release access process. That sets a pattern competitors and regulators will watch. The cyber capability could help defenders, and it also raises misuse questions if access widens.

The 'next era of frontier intelligence' framing is marketing. Every benchmark figure and internal case study (the 40% quantum subroutine gain, 300+ TiB of memory freed, the 2.7x faster libgav1 decoder) comes from Google and has not been independently verified. The case studies are curated successes with no baselines or failure rates, and the announcement links no primary sources.

1. 01Independent replication of the DeepSWE, CWE-bench and AutomationBench results by third parties, and whether the Vals Index ranking holds on Vals AI's own published pages.
2. 02A dated general availability announcement, and whether the final pricing matches the $2 and $10 per million token figures or changes.
3. 03Evidence on how the Fairwind Program is run: who gets access, what the lifted cyber guardrails allow, any reported misuse or leaks, and any public output from the US government pre-release process.