Google releases updated Gemini 3 Deep Think with record-breaking benchmark results

Google has unveiled an updated version of its Gemini 3 Deep Think model, achieving state-of-the-art performance across multiple artificial intelligence benchmarks. The model reached top results in three key technical evaluations: ARC-AGI-2, MMMU-Pro, and HLE. The system also demonstrated gold-medal-level performance in Physics and Chemistry competitions, matching the highest standards set by international academic olympiad competitions. The advancements underscore the effectiveness of extended reasoning architectures in tackling highly complex problem-solving tasks.