At Google I/O 2025, Google DeepMind CEO Demis Hassabis announced DeepThink, a new reasoning mode built into Gemini 2.5 Pro that showcases the company’s latest breakthroughs in AI logic and problem-solving.
Using advanced techniques like parallel thinking, DeepThink is designed to improve how AI handles complex, multi-step tasks across math, code, and multimodal understanding.
“So far, we’ve seen incredible performance,” said Hassabis on stage. “DeepThink uses our latest cutting-edge research in thinking and reasoning, including parallel techniques. It gets an impressive score on USAMO 2025, currently one of the hardest math benchmarks. It leads on LiveCodeBench, a difficult benchmark for competition-level coding. And since Gemini has been natively multimodal from the start, it’s no surprise that it also excels on the main benchmark measuring lists, MMMU.”
Google is making DeepThink available to a limited group of trusted testers through the Gemini API before rolling it out more broadly. Hassabis explained that the company is taking additional time to conduct frontier safety evaluations and gather feedback from safety experts to ensure responsible deployment.
“As part of that, we’re going to make it available to trusted testers via the Gemini API to get their feedback before making it widely available,” he said. “You’ll hear more about our plans for 2.5 Pro DeepThink from Josh later today.”
DeepThink is being positioned as a major milestone in AI reasoning, and Google is framing it as one of the most advanced tools it has developed to date.
Check out the ongoing Google I/O ’25 Keynote below:
Want to see more of our stories on Google?
P.S. Want to keep this site truly independent? Support us by buying us a beer, treating us to a coffee, or shopping through Amazon here. Links in this post are affiliate links, so we earn a tiny commission at no charge to you. Thanks for supporting independent Canadian media!
