Google has released Gemini 2.5 Pro, its first model in the Gemini 2.5 family and most capable reasoning model to date. The model features a 1 million token input context window (with plans for expansion), supports multimodal inputs including text, image, audio, and video, and is currently available via the Gemini Advanced plan and Google AI Studio.
- Context Window: Up to 1 million tokens for input, expanding to 2 million; output size is 64,000 tokens.
- Benchmarks: Leads GPQA Diamond (84.0%), AIME 2024 (92.0%), and MMMU (81.7%); scores 18.8% on Humanity’s Last Exam.
- Long Context: Achieves 91.5% on MRCR for 128,000 context length, significantly outperforming competitors like o3-mini.
- Coding: Scores 63.8% on SWE-bench verified and 74.0% on Aider Polyglot.
The model is designed to handle complex multi-step tasks, code generation, and large document analysis without relying on retrieval-augmented generation (RAG), offering significant utility for enterprise adoption.