Google DeepMind unveiled Gemini 2.5 on March 25, 2025, calling it their most intelligent AI model yet. This experimental version, Gemini 2.5 Pro, claims the top spot on the LMArena leaderboard and showcases improved reasoning and coding skills.
What does it offer?
Gemini 2.5 models are designed as “thinking models.” They can reason through their thoughts before responding, which boosts both performance and accuracy. This involves analyzing information, making logical deductions, understanding context, and making informed decisions.
Gemini 2.5 Pro Experimental is built for complex tasks. The model leads in common coding, math, and science benchmarks.
Gemini 2.5 Pro is currently available in Google AI Studio and the Gemini app for Gemini Advanced users. It’s slated to arrive on Vertex AI in the coming weeks, with pricing and higher rate limits for production use also on the way.
The model achieved a state-of-the-art score of 18.8% on Humanity’s Last Exam, a dataset crafted by hundreds of experts to measure human knowledge and reasoning, and leads the field in math and science benchmarks like GPQA and AIME 2025.
Gemini 2.5 Pro has made strides in coding performance. The model received 63.8% on SWE-Bench Verified with a custom agent setup, in agentic code evaluations.
Gemini 2.5 Pro retains the core strengths of Gemini models, including native multimodality and a long context window. It ships with a 1-million-token context window, with a 2-million-token version coming soon. This allows it to process large datasets and tackle problems involving text, audio, images, video, and even entire code repositories.
Developers and businesses can experiment with Gemini 2.5 Pro in Google AI Studio immediately. Gemini Advanced users can select it in the model dropdown, and Vertex AI access will follow in the coming weeks.
