On Tuesday morning, Google unveiled Gemini 2.5, which is described as its most intelligent AI model to date. The first member of Google’s new thinking AI models is an experimental version of Gemini 2.5 Pro, which currently tops the LMArena leaderboard. It is ahead of every major AI model, including Grok 3, GPT-4.5, and DeepSeek-R1.
Like other reasoning models, Google says that Gemini 2.5 Pro can “analyze information, draw logical conclusions, incorporate context and nuance, and make informed decisions.” As for the math and science tests like GPQA and AIME 2025, Gemini 2.5 Pro outperforms other models, with a record-breaking 18.8% on Humanity’s Last Exam.
Gemini 2.5 Pro is also much better at coding than 2.0, excelling at “creating visually compelling web apps and agentic code applications, along with code transformation and editing.” Google even shared a video of a game demo built by the new model:
Another upgrade for 2.5 is the context window, which is now capable of holding up to 1 million tokens (with 2 million coming soon). You can feed it enormous datasets that include text, audio, images, video, and even entire code repositories.
“Now, with Gemini 2.5, we’ve achieved a new level of performance by combining a significantly enhanced base model with improved post-training,” says Koray Kavukcuoglu, CTO of Google DeepMind, in a blog post. “Going forward, we’re building these thinking capabilities directly into all of our models, so they can handle more complex problems and support even more capable, context-aware agents.”
If you want to try Gemini 2.5 Pro Experimental, it’s available right now in Google AI Studio and the Gemini app for Gemini Advanced users, with Vertex AI access coming soon. Google says that pricing will be announced in the coming weeks.
You can learn more about Gemini 2.5 on Google’s website.
