Google has announced its most intelligent AI model yet, called Gemini 2.5 Pro. This experimental version is an advanced version of Gemini 2.5 and debuts at the top of several benchmarks, including LMArena by a significant margin.
Gemini 2.5 models are “thinking” models that can reason through their thoughts before responding, resulting in enhanced performance and accuracy. They use techniques like reinforcement learning and chain-of-thought prompting to make AI smarter and more capable of reasoning.
The new model achieves a new level of performance by combining an enhanced base model with improved post-training. It will be available now in Google AI Studio and the Gemini app for advanced users, and will soon be coming to Vertex AI.
Gemini 2.5 Pro is state-of-the-art across several benchmarks requiring advanced reasoning, including math and science tasks. It excels at creating visually compelling web apps and agentic code applications, and scores high on industry-standard benchmarks for coding performance.
The model builds on Gemini’s strengths in native multimodality and long context windows, allowing it to comprehend vast datasets and handle complex problems from different information sources. Developers can start experimenting with the new model now in Google AI Studio, and Gemini Advanced users can select it in the model dropdown.
Google aims to continue improving Gemini’s abilities at a rapid pace, making its AI more helpful and capable.
Source: https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025