Topping global benchmarks and redefining reasoning in AI, Gemini 2.5 Pro brings unprecedented accuracy, coding power, and context-aware performance
Google has introduced Gemini 2.5, its most advanced AI model to date, marking a major leap in artificial intelligence with enhanced reasoning and performance across a wide array of complex tasks. Released today as an experimental version of Gemini 2.5 Pro, the model is already topping industry benchmarks and is now available in Google AI Studio and the Gemini app for Advanced users.
Gemini 2.5 Pro Experimental tops the LMArena leaderboard. Image: Google
Described as a “thinking model,” Gemini 2.5 is engineered to reason through problems before generating a response — a capability that significantly improves accuracy, contextual understanding, and decision-making. Unlike earlier models focused mainly on pattern recognition and prediction, Gemini 2.5 emphasizes logical analysis, contextual nuance, and informed decision-making.
“We’re building these thinking capabilities directly into all of our models going forward,”
Google stated in the launch announcement.
“This enables our AI to solve more complex problems and support more context-aware, capable agents.”
The new model debuts at #1 on the LMArena leaderboard, a benchmark driven by human preferences, highlighting its superior reasoning, coding, and stylistic coherence. It also performs exceptionally well on advanced math and science benchmarks, including GPQA and AIME 2025, without relying on expensive test-time methods like majority voting.
Notably, Gemini 2.5 Pro achieved a state-of-the-art 18.8% score on Humanity’s Last Exam, a rigorous dataset built by hundreds of subject matter experts to test human-level reasoning across disciplines.
The model also scores a state-of-the-art 18.8% across models without tool use on Humanity’s Last Exam
In coding, Gemini 2.5 Pro represents a significant jump from its predecessor, Gemini 2.0. It excels in generating functional, visually appealing web apps, agentic code transformations, and code editing. It also leads on SWE-Bench Verified, the industry standard for evaluating agentic coding abilities, with a score of 63.8% using a custom agent configuration.
As a showcase of its power, Google shared an example where Gemini 2.5 Pro generates a fully executable video game from a single-line prompt — demonstrating its potential for developers, educators, and creators alike.
Pricing for Gemini 2.5 Pro will be announced in the coming weeks, with options for higher rate limits and scaled production use. The model will soon be available on Vertex AI, further expanding its reach across Google’s ecosystem. Gemini 2.5 Pro is available now in Google AI Studio and in the Gemini app for Gemini Advanced users, and will be coming to Vertex AI soon.
