Multi-domain Knowledge and Reasoning on MMLU-Pro
77.7AccuracyQwen3
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3Size=14B2026.04 | 77.7 | 2,400 | |
| Apriel-Reasoner (Ours)Size=15B2026.04 | 77.3 | 1,900 | |
| Phi-4-reasoningSize=14B2026.04 | 77.1 | 3,400 | |
| Nemotron-CascadeSize=14B2026.04 | 76.8 | 3,600 | |
| Apriel-BaseSize=15B2026.04 | 76.4 | 3,500 | |
| Apriel-Base + RLVR w/ LPSize=15B, Length Penalty (LP)=true2026.04 | 75.6 | 1,500 | |
| COACTTrained on=WebInstruct, Evaluation=Out-of-domain2026.04 | 24.17 | — | |
| Pref + EntTrained on=WebInstruct, Evaluation=Out-of-domain2026.04 | 23.45 | — | |
| EntropyTrained on=WebInstruct, Evaluation=Out-of-domain2026.04 | 22.91 | — | |
| Pref CertaintyTrained on=WebInstruct, Evaluation=Out-of-domain2026.04 | 22.18 | — | |
| RandomTrained on=WebInstruct, Evaluation=Out-of-domain2026.04 | 21.54 | — |