Google has unveiled a new generation of its core AI model, Gemini 2.5. This generation includes the Gemini 2.5 Pro Experimental model, and the company says it outperforms competing models from OpenAI, Anthropic, xAI, and DeepSeek.
Gemini 2.5 is a reasoning model that combines an improved base model with post-training technology. The new model is now available to Gemini Advanced subscribers and in Google AI Studio.
The new model’s multimodality allows it to process text, audio, images, video, and code. The model has a context window of 1 million tokens, but the company promises to soon increase it to two million. In general benchmarks, the model outperforms competitors, scoring +39 points on LMArena.
Google has focused on coding performance with the new model. The company says that Gemini 2.5 Pro is ideal for developing visually appealing web applications, as well as for transforming and editing code. Google shared a demo video that shows a version of a running dinosaur game being created with a single prompt.