MUSR (Multi-Step Reasoning) from the Open LLM Leaderboard — tests ability to reason across multiple steps in complex scenarios.
Targets chained reasoning where early mistakes propagate and compound across steps; complements single-step QA by exposing planning and intermediate-state weaknesses.
Retired by Hugging Face on March 13, 2025; retained scores are historical. Scores are often low and noisy, so small rank differences can be unstable; scenario style may not reflect task decomposition used in production agent systems.
Higher is better. Percentage accuracy.
Max score: 100
Retired by Hugging Face on March 13, 2025.
| # | Model | Score |
|---|---|---|
| 1 | 50.6 | |
| 2 | 50.3 | |
| 3 | 48.3 | |
| 4 | 47.2 | |
| 5 | 47.0 | |
| 6 | 47.0 | |
| 7 | 46.9 | |
| 8 | 46.9 | |
| 9 | 46.8 | |
| 10 | 46.8 | |
| 11 | 46.4 | |
| 12 | 46.4 | |
| 13 | 46.3 | |
| 14 | Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning) source benchpress Evidence observed Apr 13, 2026 | 46.2 |
| 15 | 46.1 | |
| 16 | 46.0 | |
| 17 | 46.0 | |
| 18 | 45.9 | |
| 19 | 45.9 | |
| 20 | 45.8 | |
| 21 | 45.8 | |
| 22 | 45.8 | |
| 23 | 45.7 | |
| 24 | 45.7 | |
| 25 | 45.7 | |
| 26 | 45.7 | |
| 27 | 45.7 | |
| 28 | 45.7 | |
| 29 | 45.7 | |
| 30 | 45.6 | |
| 31 | 45.6 | |
| 32 | 45.6 | |
| 33 | 45.5 | |
| 34 | 45.5 | |
| 35 | 45.5 | |
| 36 | 45.5 | |
| 37 | 45.4 | |
| 38 | 45.4 | |
| 39 | 45.4 | |
| 40 | 45.4 | |
| 41 | 45.4 | |
| 42 | 45.3 | |
| 43 | 45.3 | |
| 44 | 45.3 | |
| 45 | 45.2 | |
| 46 | 45.2 | |
| 47 | 45.1 | |
| 48 | 45.1 | |
| 49 | DeepHermes 3 - Mistral 24B Preview (Non-reasoning) source open_llm_leaderboard Evidence observed Sep 19, 2026 | 45.0 |
| 50 | 45.0 |