Step 3.5 Flash ranks first on CF-Div2-Stepfun.
StepFun Feb 2026 Open weights Mid-size
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 | |
|---|---|---|---|---|
| MMLU-Pro | 84.4% | #39 of 167 | Lab model card | |
| Arena-Hard v2.0 | 74% | #9 of 24 | Lab model card | |
| LMArena (Elo) | 1,393 | #139 of 335 | Official leaderboard | |
| RepoQA | 88.5% | #2 of 4 | Lab model card | |
| LongBench v2 | 57.5% | #18 of 35 | Lab model card | |
| BullshitBench | 25.4% | #58 of 107 | Official leaderboard | |
| MultiChallenge | 55.7% | #23 of 42 | Lab model card | |
| KINA | 29.1% | #23 of 37 | Official leaderboard | |
| Humanity's Last Exam | 23.1% | #96 of 154 | Lab model card | |
| IFBench | 67.4% | #50 of 80 | Lab model card | |
| BullshitBench V2 | 22% | #87 of 129 | Official leaderboard | |
| Extended NYT Connections | 28.4% | #81 of 103 | Official leaderboard | |
| FRAMES | 76.5% | #8 of 9 | Lab model card | |
| StudyArena Text | 1,124 | #85 of 85 | Official leaderboard | |
| OpenAI MRCR | 28.8% | #10 of 10 | Lab model card | |
| Consulting and Recommendations Benchmark (StepFun) | 70.5% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BRUMO 2025 | 100% | #2 of 42 | Official leaderboard | |
| CMIMC 2025 | 95.6% | #3 of 33 | Official leaderboard | |
| HMMT Nov 2025 | 94.2% | #2 of 22 | Official leaderboard | |
| HMMT 2025 | 98.4% | #5 of 54 | Lab model card | |
| AIME 2025 | 97.3% | #16 of 143 | Lab model card | |
| SMT 2025 | 91.5% | #5 of 40 | Official leaderboard | |
| AIME 2026 | 96.7% | #7 of 31 | Official leaderboard | |
| GPQA Diamond | 83.5% | #92 of 302 | Lab model card | |
| IMO-AnswerBench | 85.4% | #13 of 37 | Lab model card | |
| MathArena Apex Shortlist | 70.7% | #15 of 38 | Official leaderboard | |
| MathArena (Apex, IMO, USAMO) | 13.5% | #19 of 47 | Official leaderboard | |
| HMMT 2026 | 86.4% | #22 of 31 | Official leaderboard | |
| USAMO 2026 | 44.6% | #7 of 9 | Official leaderboard | |
| ARC-AGI | 54.8% | #17 of 21 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CF-Div2-Stepfun | 86.1% | #1 of 8 | Official leaderboard | |
| LiveCodeBench v6 | 86.4% | #17 of 90 | Lab model card | |
| SWE-bench Verified | 74.4% | #50 of 155 | Lab model card | |
| SWE-bench Multilingual | 67.4% | #48 of 73 | Lab model card | |
| Data Analysis Benchmark (StepFun) | 39.6% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BrowseComp-ZH | 73.7% | #4 of 19 | Lab model card | |
| tau2-bench | 88.2% | #19 of 79 | Lab model card | |
| LisanBench | 2409.7 | #39 of 133 | Official leaderboard | |
| PinchBench 2.0 | 79.3% | #18 of 57 | Official leaderboard | |
| xbench-DeepSearch | 83.7% | #1 of 3 | Lab model card | |
| GAIA | 84.5% | #2 of 5 | Lab model card | |
| PinchBench | 70.5% | #22 of 48 | Official leaderboard | |
| SnakeBench | 23 | #101 of 188 | Official leaderboard | |
| BrowseComp | 69% | #46 of 81 | Lab model card | |
| Terminal-Bench 2.0 | 51% | #37 of 63 | Lab model card | |
| ResearchRubrics | 65.3% | #4 of 6 | Lab model card | |
| ClawProBench | 41.8% | #48 of 49 | Official leaderboard | |
| WildClawBench | 26.7% | #31 of 31 | Official leaderboard | |
| AndroidDaily | 57% | #2 of 2 | Lab model card | |
| xbench-DeepSearch-2510 | 56.3% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveSecBench | 52.1 | #19 of 34 | Official leaderboard |
The Self-Reported Index fits one score to the 21 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 142.2 to 148.0. How we compute it.
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 |
|---|---|---|---|
| AIME 2025 | 97.3% AIME 2025 | 98.3% Leaderboard | −1.0 pts |
| HMMT Nov 2025 | 94% HMMT 2025 (Nov.) | 94.2% Leaderboard | −0.2 pts |
| HMMT 2025 | 98.4% HMMT 2025 (Feb.) | 98.3% Leaderboard | Same |
| CF-Div2-Stepfun | 86.1% CF-Div2-Stepfun-cpp, avg@8 (Tables 5 and 15). PaCoRe 93.3. Python 81.5, Java 77.1, estimated Codeforces C++ pass@8 rating 2489 (Table 15) | 86.1% Leaderboard, C++; avg@8 | Same |
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 16 of them aren't in the launch post.