Multi-table Reasoning on MMQA 3-table
40.88AccuracyCPO
Evaluation Results
| Method | Links | |
|---|---|---|
| CPOModel=Qwen3-14B2026.06 | 40.88 | |
| CPOModel=Llama-3.1-8B2026.06 | 35.19 | |
| CPOModel=Mistral-8B2026.06 | 34.32 | |
| Trace SFTModel=Llama-3.1-8B2026.06 | 28.84 | |
| Q&A SFTModel=Qwen3-14B2026.06 | 28.53 | |
| Trace SFTModel=Qwen3-14B2026.06 | 27 | |
| Trace SFTModel=Mistral-8B2026.06 | 24.85 | |
| Q&A SFTModel=Mistral-8B2026.06 | 21.47 | |
| Q&A SFTModel=Llama-3.1-8B2026.06 | 18.97 |