Google withholds Argon AI model from public release

Alphabet’s Google announced Argon on Wednesday as the anchor for its Gemini 4 generation.[1] Google framed the launch as part of its effort to catch up with AI rivals Anthropic and OpenAI.[1] The company described Argon as its most performant model yet, built for complex workloads, and said it is larger than its previous advanced Pro models.[1] Despite those capabilities, Google said it would withhold the model from the public for now.[1] Its initial access is restricted to a vetted group of cybersecurity experts, with selected cybersecurity partners also receiving access.[1]

Google’s release reported that Argon scored better than OpenAI’s Astra and Anthropic’s Opus on several industry benchmarks.[1] The results were not uniformly stronger: Google said Argon lagged on two of the four coding-related benchmarks included in its release.[1] That distinction matters for teams considering the model for software work; the evidence does not show that Argon leads on every coding measure. Google says it is suited to complex workloads in software engineering, legal and financial tasks, and cyber defence.[1] Those are stated areas of suitability, not evidence that the model is publicly available for those uses.[1]

Google says Argon has a leading ability to find and fix critical software flaws.[1] The company reported that early testers used it to uncover a flaw in software used by hospitals worldwide, exposing sensitive personal information; Google said other advanced models had missed that flaw.[1] Google also says Argon is designed to refuse requests that could facilitate cyberattacks or the development of chemical, biological or nuclear weapons.[1] The restricted rollout reflects concern among cybersecurity experts that advanced AI could be used to hack banks, hospitals and government systems.[1]

Google said safely releasing frontier capabilities at this level requires a phased approach, and it has not given a timeline for Argon’s public release.[1] The company is participating in the Trump administration’s voluntary pre-release model-access process.[1] The announcement followed a White House meeting the previous day, attended by executives including Google CEO Sundar Pichai and Anthropic CEO Dario Amodei; executives at the meeting signed a voluntary accord.[1] Google’s cautious rollout resembles Anthropic’s restriction of Claude Mythos Preview to a small number of trusted organisations.[1]