On November 18, 2025, Google released Gemini 3, which it called its most intelligent model, combining “state-of-the-art reasoning with unprecedented depth and nuance” with advanced multimodal understanding and agentic capabilities. Google described it as “the best vibe coding and agentic coding model we’ve ever built,” emphasizing planning and consistent tool use over extended workflows.
Google reported leading benchmark results, including the top position on the LMArena leaderboard at 1501 Elo, 91.9 percent on GPQA Diamond, and 81 percent on the MMMU-Pro multimodal benchmark. An enhanced Gemini 3 Deep Think variant reached 41.0 percent on Humanity’s Last Exam and what Google called “an unprecedented 45.1 percent on ARC-AGI-2.” Google highlighted that this was the first time it launched a Gemini model across the Gemini app, AI Studio, Vertex AI, and developer platforms on the same day.
Gemini 3 mattered as Google’s strongest entry in the late-2025 frontier race, reasserting its position against GPT-5 and the Claude line. Its simultaneous launch across consumer and developer products signaled that frontier capability and broad distribution had become a single, coordinated move rather than a staged rollout.