Google has finally shown what was supposed to be its big answer to OpenAI and Anthropic this year. The company unveiled Argon, the flagship model anchoring its entire Gemini 4 lineup. The release lands roughly six months behind Sundar Pichai's original timeline, and the previous flagship, Gemini 3.5 Pro, has quietly been shelved for good.
A Google spokesperson called Argon "our most performant model yet built for complex workloads," comparable on key coding and cyber benchmarks to OpenAI's Astra and Anthropic's Opus. The numbers Google itself released are more honest than the pitch: Argon beats its rivals on several benchmarks but trails on two of the four coding tests the company chose to publish.
Broad access isn't happening yet. Right now Argon is limited to select cybersecurity partners, and the model is part of the Trump administration's voluntary review process, where companies share frontier models with officials before any public launch. Google gave no date for when ordinary users might get it.
The delay tracks with a rough stretch at DeepMind: founder Demis Hassabis stepped back from running the lab, and several senior Gemini leaders left the company. While Google sorted out its own structure, OpenAI and Anthropic kept shipping updates and widening what had recently looked like an even race.
The benchmark comparisons are Google's own measurements, not an independent audit, and they're worth reading with that in mind. With Argon in the hands of a handful of partners and no public launch date attached, the real test won't come from a press release — it'll come from the first independent evaluations, whenever those actually happen.
