"About a quarter of them contained bugs": Gemini 4's ML R&D benchmark

Gemini 4's machine learning research and development benchmark revealed that about 25% of tested models contained bugs, highlighting challenges in AI model reliability and quality assurance.