Gemini Pro 2.5 Deep Think features «parallel thinking,» letting it work through tens, if not hundreds, of different solutions to a problem simultainiously, test them, revise them or even combine them to return the best answer.
For researchers, scientists, and academics tackling hard problems: Gemini 2.5 Deep Think is here. 🤯
It doesn't just answer, it brainstorms using parallel thinking and reinforcement learning techniques. We put it into the hands of mathematicians who explored what it can do ↓ pic.twitter.com/lsBq3ICvuM
— Google DeepMind (@GoogleDeepMind) August 1, 2025
Good at math and creativity
As opposed to the IMO gold model, that spent hours to come up with answers, Deep Think should also be faster and «more usable day-to-day.»
It should be excellent at math discovery, and at «work that needs creativity and strategic planning, including tackling tough coding problems as well as iterative web design.»
The model solves complex problems step-by-step, and can improve both he «aesthetics and functionality» in web development.
Impressive benchmarks
It also scores impressively well at the benchmarks, reaching a never-before-seen 34.8% on Humanity’s Last Exam (with no tools), beating the previous record by Grok 4 with almost 10%.
This should be enough to call Deep Think the best model out there, and certainly qualifies for state of the art status.
It also scores very nicely in LiveCodeBench, AIME 2025 and IMO 2025 — where it almost doubles the previous top score.
The model is only available to users on the $250 Gemini Ultra plan, and even then it’s rare and expensive — users will only get «a few prompts per day,» writes 9to5Google.
With this, that’s X.ai out with a next generation mode, and now Google with theirs — so now we wait for Anthropic and OpenAI to respond. GPT 5 should be out any day now.
Read more: Google’s launch post, writeups on 9to5Google, TechCrunch and Ars Technica.