Google just introduced DeepThink in Gemini 2.5 Pro!! (benchmarks are wild 🤯)

Google finally did it. They added what may as well be the finishing touch to their already superior Gemini 2.5 Pro model, they added a DeepThink functionality.

We’re making 2.5 Pro even better by introducing Deep Think — an enhanced reasoning mode that uses our latest cutting-edge research in reasoning, including parallel thinking techniques. And so far we’ve seen incredible performance.

Gemini 2.5 Pro DeepThink Performance Benchmarks

Google stated the following on X about the benchmarks:

Gets an impressive score on USAMO 2025, currently one of the hardest math benchmarks

Leads on LiveCodeBench, a difficult benchmark for competition-level coding

And excels on MMMU, which tests multimodal reasoning

We’re making it available to trusted testers via the Gemini API to get their feedback before making it widely available.

How to use DeepThink

While DeepThink for Gemini 2.5 Pro is not yet available for public / API use, you can test it in "Jules", which is an asynchronous coding agent from Google that can tackle complex tasks in large codebases that used to take hours. It can plan steps, modify files and more in just minutes. ⏱️

We are personally very excited to see how it will perform in coding editors like Cursor or Windsurf.

Want to test DeepThink? You'll have to use Googles new "Jules": http://jules.google