Legal Question Answering on LEXam Overall
71.9AccuracyEP-HUBO (Qwen-35B)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| EP-HUBO (Qwen-35B)Adjudicator=Opus 4.6, Trace Generator=Qwen3.5-35B, Number of Traces=202026.06 | 71.9 | 5.1 | 68.8 | 53 | 24 | |
| EP-HUBO (OSS-20B)Adjudicator=Opus 4.6, Trace Generator=OSS-20B, Number of Traces=202026.06 | 71.5 | 4.7 | 64.5 | 60 | 33 | |
| MV (Qwen-35B)Trace Generator=Qwen3.5-35B, Number of Traces=20, Evaluation Protocol=Majority Vote2026.06 | 69.1 | — | — | — | — | |
| ZS (no traces/HUBO)Adjudicator=Opus 4.6, Evaluation Protocol=Zero-shot2026.06 | 66.8 | — | — | — | — | |
| MV (OSS-20B)Trace Generator=OSS-20B, Number of Traces=20, Evaluation Protocol=Majority Vote2026.06 | 48.3 | — | — | — | — | |
| EP-HUBO (Qwen-35B)Adjudicator=Sonnet 4.6, Trace Generator=Qwen3.5-35B, Number of Traces=202026.06 | 42.5 | 20.2 | 92 | 126 | 11 | |
| EP-HUBO (OSS-20B)Adjudicator=Sonnet 4.6, Trace Generator=OSS-20B, Number of Traces=202026.06 | 33.7 | 11.4 | 80.4 | 86 | 21 | |
| ZS (no traces/HUBO)Adjudicator=Sonnet 4.6, Evaluation Protocol=Zero-shot2026.06 | 22.3 | — | — | — | — |