We are proud to announce the launch of Gemini 2.5, the most advanced AI model in our series. The initial 2.5 release is an experimental version of 2.5 Pro, which has surpassed existing benchmarks and secured the top spot on the LMArena leaderboard by a considerable margin.

The Gemini 2.5 models are designed to think critically, allowing them to analyze their thoughts before providing a response, resulting in improved performance and increased accuracy.

In the realm of artificial intelligence, a system’s ability to “reason” encompasses more than just classification and prediction. It involves the capacity to analyze information, draw logical conclusions, incorporate context and subtlety, and make well-informed decisions.

For an extended period, we have been exploring methods to enhance AI intelligence and reasoning capabilities through techniques such as reinforcement learning and chain-of-thought prompting. Recently, we introduced our first thinking model, Gemini 2.0 Flash Thinking, building upon these advancements.

With the introduction of Gemini 2.5, we have achieved a new level of performance by combining a significantly enhanced base model with improved post-training techniques. Moving forward, we will be integrating these thinking capabilities into all our models, enabling them to tackle more complex problems and support more advanced, context-aware agents.