Google has announced Gemini 2.5 Pro Experimental, its most powerful and intelligent AI model to date. The new model features enhanced reasoning, deeper contextual understanding, and improved logical inference, making it more accurate and reliable across a wide range of tasks.
Gemini 2.5 leads in academic and scientific benchmarks, including top scores in GPQA and AIME 2025, and excels in programming with a verified 63.8% score on SWE-Bench — thanks to improved agent capabilities for editing and generating complex applications.
One of the model’s standout features is its extended context window: it can now process up to 1 million tokens, with plans to scale up to 2 million soon. This allows Gemini 2.5 to work across massive amounts of data, including large codebases, documents, audio, images, and video.
Gemini 2.5 is already available to developers through Google AI Studio and to Gemini Advanced subscribers. Enterprise integration via Vertex AI is also underway. With this release, Google aims to pave the way for more intelligent AI agents capable of understanding and navigating rich, complex contexts.