Gemini 2.5 Deep Think
Google DeepMind · 2025-08-01
Gemini 2.5 Deep Think is Google's enhanced reasoning model in the Gemini 2.5 family, released to Google AI Ultra subscribers in the Gemini app on August 1, 2025 as a Deep Think toggle under 2.5 Pro. It combines parallel thinking with additional reinforcement learning to test multiple hypotheses at once, and the app release was a faster variation of the full competition model that reached gold-medal standard at the 2025 International Mathematical Olympiad. At launch it had no separate public API identifier; Google said API access with and without tools was planned for trusted testers.
Benchmark scores
| Benchmark | Score |
|---|---|
| AIME 2025 | 99.2 |
| Capability | 154.8 |
| Humanity's Last Exam | 34.8 |
| IMC 2025 | 93.0 |
| IMO 2025 | 60.7 |
| LiveCodeBench | 87.6 |
| VoxelBench Text Rating | 1491 |