Gemini 2.5 Pro May 2025
Gemini 2.5 Pro ranks first on VideoReasonBench.
35
Benchmarks with a score
1
Ranked first, out of five or more models
6
In the top three
Google DeepMind May 2025 Gemini 2.5 Pro
May 2025 snapshot Closed weights
140.9
#119 of 256
Epoch Capabilities Index
- Humanity's Last Exam 17.8% #24 /47
- GPQA Diamond 66.7% #125 /229
- VideoReasonBench 56% #1 /14
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 | |
|---|---|---|---|---|
| LLM Confabulation Leaderboard | 10.6% | #2 of 58 | Official leaderboard | |
| ECLeKTic | 40.1% | #2 of 35 | Official leaderboard | |
| AraGen | 68.4% | #13 of 42 | Official leaderboard | |
| Fiction.LiveBench | 66.7% | #25 of 57 | Fiction.live leaderboard, via Epoch AI | |
| EQ-Bench 3 | 1,271 | #39 of 78 | Official leaderboard | |
| Humanity's Last Exam | 17.8% | #24 of 47 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MATH Level 5 | 95.9% | #10 of 94 | Epoch AI run | |
| Misguided Attention | 58.8% | #7 of 45 | Official leaderboard | |
| HMMT 2025 | 80.8% | #23 of 53 | Official leaderboard | |
| GPQA Diamond | 66.7% | #125 of 229 | Epoch AI run | |
| AIME 2025 | 83.3% | #30 of 53 | Official leaderboard | |
| BRUMO 2025 | 89.2% | #26 of 42 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Aider Polyglot | 76.9% | #6 of 46 | Official leaderboard | |
| LiveCodeBench v5 | 73.7% | #5 of 21 | Official leaderboard | |
| LiveCodeBench v6 | 71.8% | #5 of 21 | Official leaderboard | |
| LiveCodeBench | 87.5% | #6 of 21 | Official leaderboard | |
| WebDev Arena | 1,420 | #2 of 2 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AceBench | 80.6% | #3 of 16 | Official leaderboard | |
| lmgame-Bench | 42.0 | #6 of 23 | Official leaderboard | |
| GAIA | 73.1% | #17 of 64 | Official leaderboard | |
| Berkeley Function Calling Leaderboard (BFCL) | 59.9% | #24 of 67 | Official leaderboard | |
| LLM Chess | -71 | #72 of 123 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| xbench-Marketing | 45.9 | #4 of 7 | Official leaderboard | |
| xbench-Recruitment | 57.3 | #5 of 8 | Official leaderboard | |
| TaxCalcBench | 32.4% | #11 of 17 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GeoBench | 86% | #3 of 48 | Official leaderboard | |
| MMMU | 79.6% | #9 of 136 | Official leaderboard | |
| VideoReasonBench | 56% | #1 of 14 | Official leaderboard | |
| VISTA | 50.8% | #9 of 54 | Official leaderboard | |
| Point-Bench | 62.8% | #4 of 17 | Official leaderboard | |
| MMMU-Pro | 68% | #17 of 57 | Official leaderboard | |
| Video-MME | 84.8% | #21 of 51 | Lab model card | |
| VPCT | 40.5% | #13 of 27 | VPCT leaderboard, via Epoch AI |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AIRTBench | 34.3% | #3 of 9 | Official leaderboard | |
| Molecular Biology Capabilities Test | 52.2% | #31 of 90 | Official leaderboard |
Google DeepMind's models over time
Google DeepMind's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.