STEP3-VL-10B ranks first on MMBench.
StepFun Jan 2026 Open weights Small
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 | |
|---|---|---|---|---|
| WildBench | 86.0 | #1 of 4 | Lab model card | |
| MultiChallenge | 62.6% | #14 of 42 | Lab model card | |
| MM-IFEval | 61.9% | #1 of 3 | Lab model card | |
| HallusionBench | 64.9% | #13 of 31 | Lab model card | |
| MMLU-Pro | 76% | #105 of 167 | Lab model card | |
| Arena-Hard v2.0 | 58.6% | #17 of 24 | Lab model card | |
| IFEval | 82.2% | #98 of 135 | Lab model card | |
| SimpleVQA | 53.1% | #25 of 30 | Lab model card | |
| SuperGPQA | 50.4% | #30 of 35 | Lab model card | |
| IFBench | 43.3% | #70 of 80 | Lab model card | |
| Humanity's Last Exam | 10.7% | #137 of 154 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AIME 2024 | 90.9% | #16 of 88 | Lab model card | |
| AIME 2025 | 87.7% | #63 of 143 | Lab model card | |
| GPQA Diamond | 70.8% | #179 of 302 | Lab model card | |
| Beyond AIME | 63.2% | #9 of 14 | Lab model card | |
| CNMO 2024 | 78.2% | #5 of 7 | Lab model card | |
| HMMT 2025 | 78.2% | #39 of 54 | Lab model card | |
| IMO-AnswerBench | 62.1% | #37 of 37 | Lab model card | |
| PhyX | 59.5% | #3 of 3 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench | 75.8% | #12 of 54 | Lab model card | |
| LiveCodeBench v6 | 75.8% | #42 of 90 | Lab model card | |
| Design2Code | 79.5 | #2 of 2 | Lab model card | |
| HumanEval-V | 66% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ScreenSpot-v2 | 92.6% | #8 of 14 | Lab model card | |
| OSWorld-G | 59% | #3 of 5 | Lab model card | |
| GameHorizon | 60.6 | #32 of 37 | Official leaderboard | |
| ScreenSpot-Pro | 51.5% | #28 of 32 | Lab model card | |
| MMBench-GUI | 81.5% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HealthBench | 44.7% | #30 of 31 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMBench | 92% | #1 of 11 | Lab model card | |
| ReMI | 67.3% | #1 of 4 | Lab model card | |
| MMMU | 78.1% | #26 of 103 | Lab model card | |
| MathVista | 84% | #22 of 70 | Lab model card | |
| MM-MT-Bench | 8.14 | #2 of 6 | Lab model card | |
| BLINK | 66.8% | #11 of 29 | Lab model card | |
| AI2D | 89.3% | #20 of 52 | Lab model card | |
| MathVerse | 75.7% | #5 of 12 | Lab model card | |
| LogicVista | 66.9% | #6 of 14 | Lab model card | |
| MMStar | 77.5% | #16 of 36 | Lab model card | |
| OCRBench | 86.8% | #16 of 34 | Lab model card | |
| All-Angles Bench | 57.2% | #3 of 6 | Lab model card | |
| MindCube | 62.8% | #2 of 4 | Lab model card | |
| MATH-Vision (MathVision) | 70.8% | #30 of 51 | Lab model card | |
| We-Math | 73% | #8 of 13 | Lab model card | |
| RealWorldQA | 74.4% | #25 of 38 | Lab model card | |
| OmniSpatial | 51.6% | #4 of 6 | Lab model card | |
| MMMU-Pro | 64.1% | #61 of 90 | Lab model card | |
| DynaMath | 56.4% | #16 of 23 | Lab model card | |
| CVBench | 83.5% | #11 of 15 | Lab model card | |
| CountBench | 88.8% | #21 of 28 | Lab model card | |
| ERQA | 48.9% | #29 of 36 | Lab model card | |
| MMSI-Bench | 32.2% | #5 of 6 | Lab model card | |
| CharXiv | 59.5% | #64 of 75 | Lab model card | |
| V* | 82.8% | #16 of 17 | Lab model card | |
| ZeroBench | 1.00% | #31 of 31 | Lab model card | |
| CC-OCR-Bench-V2 | 41% | #15 of 15 | Official leaderboard | |
| VisuLogic | 29.7% | #7 of 7 | Lab model card | |
| MMVP | 68.2% | #5 of 5 | Lab model card | |
| CountQA | 33.7% | #3 of 3 | Lab model card | |
| ViewSpatial-Bench | 46.1% | #3 of 3 | Lab model card | |
| MIA-Bench | 92% | #2 of 2 | Lab model card | |
| PixMo-Count | 70.8% | #2 of 2 | Lab model card | |
| CoreCognition | 66.7% | #1 of 1 | Lab model card | |
| CSVQA | 63.3% | #1 of 1 | Lab model card | |
| Do You See Me | 67.5% | #1 of 1 | Lab model card | |
| M3GIA | 78.4% | #1 of 1 | Lab model card | |
| OmniAI OCR Benchmark | 77% | #1 of 1 | Lab model card | |
| SpatialViz-Bench | 45.5% | #1 of 1 | Lab model card | |
| STARE | 61.8% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveBench | 69.7% | #12 of 27 | Lab model card |
The Self-Reported Index fits one score to the 38 capability results StepFun reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 134.7 to 137.9. How we compute it.
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 52 of them aren't in the launch post.