Google's AI Ambitions on the Line with Gemini 4 Rollout

Google has launched its long-awaited flagship artificial intelligence model, Gemini 4 Argon, but internal skepticism surrounds its performance in key areas, such as coding. The company unveiled the model to a small group of trusted cybersecurity partners and plans to expand access after additional testing, starting with paid subscribers.

Gemini 4 posted leading scores on several benchmark tests, including beating OpenAI's Astra model on one that measures security skills. However, some insiders claim these metrics don't tell the whole story, as the model struggles to handle certain coding tasks when employees put it to work.

The company has labored to develop models that can compete with OpenAI and Anthropic PBC, having abandoned a planned release of Gemini 3.5 Pro in June. Google shares rose about 1.7% in extended trading after the announcement but pared earlier gains after Bloomberg reported employee skepticism about the model.

The company's coding abilities are uneven, according to people familiar with the model's internal evaluations. Gemini isn't particularly adept at front-end design, and its size is a concern, as big models can be expensive to run and potentially put pressure on Google's margins.