Google has released an experimental version of Gemini 2.5 Pro, its most advanced thinking model designed for complex tasks and enhanced reasoning capabilities. The model debuts at the top of the LMArena leaderboard by a significant margin and demonstrates state-of-the-art performance across various benchmarks.
- Tops LMArena leaderboard with high-quality style and strong reasoning.
- Leads math and science benchmarks like GPQA and AIME 2025 without test-time techniques.
- Scores 18.8% on Humanity’s Last Exam, capturing the human frontier of knowledge.
- Achieves 63.8% on SWE-Bench Verified for agentic code evaluation.
- Features a 1 million token context window with native multimodality support.
The model is now available in Google AI Studio and the Gemini app for Advanced users, with Vertex AI integration and pricing details expected in the coming weeks.