Legal Reasoning on LegalBench
79.3Balanced AccuracyLlama3.1-70B
Evaluation results
| Method | Links | |
|---|---|---|
| Llama3.1-70BPost-Training Stage=SFT2026.01 | 79.3 | |
| Llama3.1-70BPost-Training Stage=DPO2026.01 | 79.3 | |
| SaulLMParameters=141B2026.01 | 79.3 | |
| Llama3.1-70BPost-Training Stage=OOB2026.01 | 78 | |
| Nemotron1.5-49BPost-Training Stage=DPO2026.01 | 76.4 | |
| Nemotron1.5-49BPost-Training Stage=SFT2026.01 | 75.9 | |
| Qwen3-30BPost-Training Stage=DPO2026.01 | 75.4 | |
| Qwen3-30BPost-Training Stage=SFT2026.01 | 75.3 | |
| Nemotron1.5-49BPost-Training Stage=OOB2026.01 | 74.2 | |
| Qwen3-30BPost-Training Stage=OOB2026.01 | 74.1 | |
| TRACE-LFBackbone=LLaMA3-8B2026.05 | 59.26 | |
| TRACE-CSBackbone=LLaMA3-8B2026.05 | 59.26 | |
| Sequential LoRABackbone=LLaMA3-8B2026.05 | 56.66 | |
| DMTBackbone=LLaMA3-8B2026.05 | 51.42 | |
| Sequential Fine-tuningBackbone=LLaMA3-8B2026.05 | 50.7 | |
| Joint Fine-tuningBackbone=LLaMA3-8B2026.05 | 48.59 |