Hyderabad: Google has introduced Gemini 2.5, its most intelligenet AI model-- released as an experimental version of 2.5 Pro. The company has already made 2.5 Pro available with a 1 million token context window, which will be upgraded to 2 million soon.
Google says that Gemini 2.5 Pro can comprehend vast datasets and handle complex problems from different information sources, including text, audio, images, video and even entire code repositories. Gemini 2.5 models are thinking models that are supposed to have the capacity to reason through their thoughts before responding, resulting in enhanced performance and improved accuracy.
"For a long time, we’ve explored ways of making AI smarter and more capable of reasoning through techniques like reinforcement learning and chain-of-thought prompting," said Koray Kavukcuoglu, CTO of Google DeepMind. "Building on this, we recently introduced our first thinking model, Gemini 2.0 Flash Thinking."
According to Google, Gemini 2.5 represents a new level of performance achieved by integrating a significantly enhanced base model with improved post-training. The company plans to build these thinking capabilities into all their models, enabling them to tackle more complex problems and support highly capable, context-aware agents.
Gemini 2.5 Pro Experimental: Everything we know
Gemini 2.5 Pro Experimental is Google's most advanced model for complex tasks. It tops the LMArena leaderboard by a significant margin, leaving behind behemoths like OpenAI o3-mini, OpenAI GPT-4.5, Claude 3.7 Sonnet, Grok 3 Beta, and DeepSeek R1. The Gemini 2.5 Pro Experimental shows strong reasoning and code capabilities on the benchmark, leading in common coding, mathematics, and science.
Gemini 2.5 Pro excels in advanced reasoning benchmarks, outperforming in math and science tests like GPQA and AIME 2025, and achieving 18.8 per cent on Humanity's Last Exam without tool use. It also showcases improved coding performance over its predecessor, excelling in web apps, agentic code tasks, and achieving 63.8 per cent on SWE-Bench Verified. It can even create video games from simple prompts using advanced reasoning.
"Our first 2.5 model, Gemini 2.5 Pro Experimental, is a state-of-the-art thinking model, leading in a wide range of benchmarks – with impressive improvements in enhanced reasoning and coding," said Alphabet CEO Sundar Pichai.
Gemini 2.5 Pro is available now in Google AI Studio and in the Gemini app for Gemini Advanced users. It will be coming to Vertex AI soon. Google will announce the pricing of the Gemini 2.5 Pro in the coming weeks, enabling people to use the new model with higher rate limits for scaled production use.
