In a blog post on Tuesday, the company stated that Gemini 2.5 Pro, an experimental version, is “state-of-the-art across a wide range of benchmarks” and ranks #1 on LLM Arena by a significant margin.
Google describes Gemini 2.5 as a “thinking model”, capable of reasoning through its processes before responding, leading to enhanced accuracy. The company claims the model can analyse information, draw logical conclusions, and tackle complex problems more effectively.
“Without test-time techniques that increase costs, such as majority voting, 2.5 Pro outperforms in mathematics and science benchmarks, including GPQA and AIME 2025,” Google stated.
In coding, Google asserts that Gemini 2.5 Pro makes a “big leap over 2.0”, excelling in web app development, agentic code applications, and code transformation. On SWE-Bench Verified, a key benchmark for AI-generated code, it scores 63.8% with a customised agent setup.
The model retains Gemini’s core features, including multimodal capabilities and a 1 million token context window (set to expand to 2 million tokens soon). This allows it to process extensive datasets across text, audio, images, video, and code repositories.
Discover the stories of your interest
Gemini 2.5 Pro is now available in Google AI Studio and the Gemini app for Advanced users, with plans to launch on Vertex AI. Pricing details are expected to be announced in the coming weeks.
