Google's Gemini 4 Faces Internal Questions About Coding Capabilities as Rival Models Advance

Google has launched Gemini 4 (codenamed Argon) amid internal employee skepticism about the model's real-world coding performance, even as the company claims it outperforms OpenAI's GPT-6 Astra on multiple benchmarks. The model rolled out to a limited group of trusted cybersecurity defenders as part of the U.S. government's voluntary pre-release program on September 30, marking a significant moment for Google following a gap during which competitors Anthropic and OpenAI released increasingly capable frontier models.