Google Announces Gemini 2.5, an AI Model with Improved Capabilities

Google has unveiled a new version of its AI model, Gemini 2.5. The first experimental version, Gemini 2.5 Pro, demonstrates improved performance compared to previous developments, including higher results in logical thinking and code generation tests. The model belongs to the category of “thinking” AI, capable of analyzing information, taking into account context, and making logical conclusions before forming an answer. The developers note that the new version combines an improved basic architecture with improved post-training methods.

According to Google, Gemini 2.5 Pro shows leading results in tests that assess user preferences, as well as tests in mathematics and natural sciences (GPQA and AIME 2025). In the area of ​​code generation, the model demonstrates improved work with web applications and code editing. The model also showed a result of 18.8% in the so-called “Final Exam of Humanity” - a comprehensive test developed by experts to assess the depth of knowledge.

Gemini 2.5 retains key features from previous versions, including multimodality (support for text, audio, images, and video) and an expanded context window of 1 million tokens, with plans to increase to 2 million. This allows the model to handle significant amounts of data. Gemini 2.5 Pro is currently available in Google AI Studio for developers and in the Gemini Advanced app for subscribers, and is expected to be available in Vertex AI in the coming weeks. The company also plans to release pricing information for the new model soon.