Google rolls out Gemini 4 Argon to partners

Google's Gemini 4 Argon arrives as the company's long-awaited flagship model, and the rollout is deliberately cautious: it went first to a small group of trusted cybersecurity partners, with broader access promised only after additional testing, starting with paid subscribers. That staged approach sits alongside a public claim of strong benchmark results, including leading scores and a win over OpenAI's Astra model on a security-skills test.

Inside the company, though, the picture is more mixed. People with direct access to the effort say the model does less well when employees actually put it to work, and that it struggles with certain coding tasks. Their point is not that the benchmarks are wrong, but that those metrics do not tell the whole story of how the model performs in practice.

The test ahead is whether the gap between benchmark scores and day-to-day work narrows as testing continues and access widens to paying users. For Google, the stakes may lie in whether early partners and subscribers judge the model on its headline scores or on the coding tasks where insiders say it falls short.