Step 5 Preview: benchmark results
Provider: StepFun. Released 2026-09-20. Access: API.
Aggregate Rating 1819 ± 11, rank #54 of 1636 rated models, from 269 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EnterpriseOps-Gym-AA - Email | 70.19 | Task Success (%) | 100 |
| AA Long Context Reasoning | 88.33 | Accuracy (%) | 99.8 |
| AI Chess Leaderboard (Reasoning) | 1800 | Elo | 97.6 |
| Agent Arena - Tool Hallucination | -0.5 | Tool Hallucination Effect (percentage points vs baseline; lo | 95 |
| FrontierCS 2 - OCR Distillation Trajectory Sampling | 85.45 | Task score (0–100) | 94.1 |
| AA GDP.pdf - Legal - Criterion Pass Rate | 84.93 | Criterion Pass Rate (%) | 93.9 |
| AA Humanity's Last Exam | 46.48 | Accuracy (%) | 93.5 |
| Chatbot Arena (Text - Chinese) | 1521 | Arena Score | 92.9 |
| AA SciCode | 58.91 | Accuracy (%) | 92.7 |
| Epoch AI - Scicode | 58.91 | Score | 92.5 |
| SEAL - SWE Atlas - Codebase QnA | 63.6 | Score | 90.6 |
| ReLE - Law - Lawyer Qualification Exam | 81.3 | Accuracy (%) | 90.5 |
Interactive version: theaggregate.ai/model?slug=step-5-preview · How It Works · Data refreshed daily, snapshot 2026-11-06.