Sat Mar 29 12:30:00 UTC 2025: ## Google Unveils Gemini 2.5: A Giant Leap in AI Reasoning
**Mountain View, CA –** Google today announced Gemini 2.5, its most advanced AI model to date, boasting significantly improved reasoning and coding capabilities. The experimental version, Gemini 2.5 Pro Experimental, has already achieved top rankings on several industry benchmarks, including a substantial lead on LMArena, a measure of human preference.
Gemini 2.5 is built upon Google’s “thinking model” architecture, which allows the AI to analyze information, draw conclusions, and incorporate context before generating responses, leading to enhanced accuracy. This builds on the foundation laid by Gemini 2.0 Flash Thinking and utilizes advanced post-training techniques for optimized performance.
The new model excels in complex tasks, achieving state-of-the-art results across a range of benchmarks requiring advanced reasoning, including math, science, and coding. Specifically, Gemini 2.5 Pro Experimental scored a remarkable 63.8% on SWE-Bench Verified, an industry standard for evaluating agentic code. It also demonstrates impressive capabilities in generating executable code from simple prompts, as exemplified by its ability to create a video game from a single-line instruction.
Beyond its performance on specific benchmarks, Gemini 2.5 Pro boasts a 1 million token context window (expanding to 2 million soon), enabling it to process vast amounts of data from diverse sources including text, audio, images, video, and code repositories.
Gemini 2.5 Pro Experimental is currently available in Google AI Studio and the Gemini app for Gemini Advanced users. It will soon be accessible through Vertex AI, with pricing details to be released in the coming weeks. Google encourages user feedback to further refine and improve the model’s capabilities.
First Piper 