Olmo 3 32B Think places in the top three on 5 benchmarks.
Ai2 Nov 2025 Open weights Flagship
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.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AGIEval | 88.2% | #2 of 35 | Model card | |
| BIG-Bench Hard | 89.8% | #5 of 71 | Model card | |
| AlpacaEval 2.0 | 74.2% | #2 of 16 | Model card | |
| IFEval | 89% | #38 of 126 | Model card | |
| MMLU | 85.4% | #52 of 127 | Model card | |
| PopQA | 31.9% | #3 of 7 | Model card | |
| LMArena (Elo) | 1,306 | #222 of 333 | Official leaderboard | |
| IFBench | 47.6% | #62 of 70 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MATH | 96.1% | #3 of 93 | Model card | |
| AIME 2024 | 76.8% | #48 of 87 | Model card | |
| GPQA Diamond | 58.1% | #212 of 284 | Model card | |
| AIME 2025 | 72.5% | #107 of 143 | Model card | |
| ZebraLogic | 76% | #20 of 23 | Model card | |
| OMEGA | 50.8% | #2 of 2 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench | 83.5% | #5 of 73 | Model card | |
| HumanEval+ | 91.4% | #8 of 27 | Model card | |
| MBPP+ | 68% | #13 of 15 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LisanBench | 333.7 | #78 of 133 | Official leaderboard | |
| SnakeBench | 19 | #137 of 183 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BBQ (Bias Benchmark for QA) | 88.2% | #8 of 24 | Model card | |
| WildGuardTest | 89.4% | #2 of 5 | Model card | |
| HarmBench | 73.5% | #2 of 4 | Model card | |
| StrongREJECT | 80.8% | #14 of 22 | Model card | |
| WMDP | 34.8% | #2 of 3 | Model card | |
| XSTest | 93.9% | #12 of 16 | Model card | |
| Tulu 3 Safety Eval | 68.8% | #6 of 6 | Model card | |
| DoAnythingNow (DAN) | 20.2% | #2 of 2 | Model card | |
| TrustLLM JailbreakTrigger | 73.3% | #2 of 2 | Model card |
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 10 of them aren't in the launch post.