Google's Gemini 4 Argon Rolls Out, but Some Inside Doubt Its Coding Skills

Google's parent company Alphabet has started rolling out its flagship AI model Gemini 4 Argon, first to selected cybersecurity partners and paid subscribers, with wider access planned later. Google says the model scores high on many benchmarks, even beating OpenAI's Astra on a safety test.

But some staff working on the project say the scores don't tell the whole story. The model struggles with some coding tasks, especially front-end design, which could hurt Google in the competitive AI coding tools market. Google denies this.

People familiar with the matter say Gemini 4 is very large and costly to run, which may pressure profits. Experts say Google may be caught in the industry habit of chasing benchmark scores. Google had already dropped its Gemini 3.5 Pro release.

After the news, Alphabet shares rose about 1.7% after hours, but gains faded during regular trading as doubts surfaced.