Step3 model card (Hugging Face) Jul 31, 2025
- MMMU 74.2
- MATH-Vision 64.8
- SimpleVQA 62.2
- Hallusion Bench 64.2
- ZeroBench (sub) 23
- DYNMATH 50.1
- AIME25 82.9
- GPQA-Diamond 73
- LiveCodeBench (202408-202505) 67.1
- HMMT25 70
- CNMO24 83.7
Step 3
Step 3 places in the top three on CNMO 2024.
18
Benchmarks with a score
0
Ranked first, out of five or more models
1
In the top three
StepFun Jul 2025 Step 3
Open weights Flagship
135.6
#224 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- SimpleVQA 62.2% #17 /30
- HallusionBench 64.2% #15 /31
- GSM8K-V 25.9% #15 /24
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 | |
|---|---|---|---|---|
| HallusionBench | 64.2% | #15 of 31 | Lab model card | |
| LMArena (Elo) | 1,349 | #179 of 335 | Official leaderboard | |
| SimpleVQA | 62.2% | #17 of 30 | Lab model card | |
| ConStory-Bench | 0.84 | #10 of 14 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CNMO 2024 | 83.7% | #3 of 7 | Lab model card | |
| GPQA Diamond | 73% | #162 of 302 | Lab model card | |
| AIME 2025 | 82.9% | #83 of 143 | Lab model card | |
| HMMT 2025 | 70% | #48 of 54 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v6 | 67.1% | #64 of 90 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnakeBench | 18 | #157 of 188 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU | 74.2% | #49 of 103 | Lab model card | |
| ZeroBench | 23% | #18 of 31 | Lab model card | |
| GSM8K-V | 25.9% | #15 of 24 | Official leaderboard | |
| MATH-Vision (MathVision) | 64.8% | #35 of 51 | Lab model card | |
| Vision Arena | 1,144 | #103 of 136 | Official leaderboard | |
| DynaMath | 50.1% | #19 of 23 | Lab model card | |
| BabyVision | 14.7% | #26 of 27 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveSecBench | 38.9 | #21 of 34 | Official leaderboard |
Its lab's numbers put it at 135.6, number 224 of 404
The Self-Reported Index fits one score to the 10 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 133.1 to 138.0. How we compute it.
- AIME 2025 82.9
- MMMU 74.2
- GPQA Diamond 73
- HMMT 2025 70
- LiveCodeBench v6 67.1
- MATH-Vision (MathVision) 64.8
- HallusionBench 64.2
- SimpleVQA 62.2
- DynaMath 50.1
- ZeroBench 23
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.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| MATH-Vision (MathVision) | 64.8% MATH-Vision | 64.8% Leaderboard | Same |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.