- Humanity's Last Exam (without tools) 48.4
- ARC-AGI-2 (verified by the ARC Prize Foundation) 84.6
- Codeforces Elo 3455
- International Math Olympiad 2025, gold-medal level
- 2025 International Physics Olympiad written sections, gold medal-level
- 2025 Chemistry Olympiad written sections, gold medal-level
- CMT-Benchmark 50.5
- MMMU-Pro, no tools, 81.5% 81.5
Gemini 3 Deep Think
Gemini 3 Deep Think places in the top three on 2 benchmarks.
8
Benchmarks with a score
0
Ranked first, out of five or more models
2
In the top three
Google DeepMind Feb 2026 Gemini 3 Deep Think
Closed weights
No Epoch Capabilities Index score yet
- MMMU-Pro 81.5% #12 /90
- ARC-AGI-2 84.6% #14 /98
- ARC-AGI 96% #17 /95
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Humanity's Last Exam | 48.4% | #46 of 154 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| IMO-ProofBench | 76.7% | #2 of 20 | Official leaderboard | |
| ARC-AGI-2 | 84.6% | #14 of 98 | Official leaderboard | |
| ARC-AGI | 96% | #17 of 95 | Official leaderboard | |
| CMT-Benchmark | 50.5% | #1 of 2 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Codeforces rating | 3,455 | #2 of 47 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU-Pro | 81.5% | #12 of 90 | Lab model card | |
| VoxelBench | 1,446 | #40 of 56 | Official leaderboard |
Its lab's number next to a leaderboard
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.