Google releases new inference AI model Gemini 2.5 and 2.5 Pro experimental version

Google released the latest artificial intelligence inference model Gemini2.5 on the developer platform Google AI Studio and Gemini applications. This model has the ability to stop and "think", marking a new advance in artificial intelligence technology. At the same time, Google also launched Gemini2.5Pro Experimental, a multi-modal reasoning AI model that is considered the smartest to date.

The launch of Gemini2.5 is a new service from Google for Gemini Advanced subscribers of $20 a month, and users can experience its powerful inference capabilities. Since OpenAI released its first AI inference model in September 2024, major technology companies such as Anthropic, DeepSeek, Google and xAI have launched their respective inference models in order to compete with each other at the level of intelligence.

Google claims that Gemini2.5Pro surpasses its previous AI models and some of its leading competitors in multiple benchmarks. For example, in a code editing evaluation called Aider Polyglot, Gemini2.5Pro scored 68.6%, surpassing the top AI models of OpenAI, Anthropic and DeepSeek.

In addition, in the "Last Test for Humanity", Gemini2.5Pro scored 18.8%, which also outperformed many competitors. Google says the Gemini2.5Pro comes with a context window of 1 million tokens, which allows the AI ​​model to process about 750,000 words in a single prompt, even longer than the entire Lord of the Rings series. In the future, Google plans to launch a context window of 2 million tokens to further improve the capabilities of the model.

Official blog: https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/#advanced-coding