Analysis: Google launches Gemini 4 to catch up with AI rivals, but internal views on its coding performance are mixed

PANews, October 1 - Google has begun rolling out its highly anticipated flagship AI model, Gemini 4 Argon, but there are doubts within the company about the model's actual performance in key areas such as coding. On Wednesday, Google opened the model to a small group of trusted cybersecurity partners and said it would expand access after further testing, starting with paying users. Gemini 4 achieved leading results on multiple benchmarks, including beating OpenAI's Astra model on a test measuring safety capabilities.

However, some insiders say these metrics do not tell the whole story. According to people directly familiar with the work, although Gemini 4 performs well on benchmarks measuring model effectiveness, it is less impressive when employees actually use it, especially for certain coding tasks.