Long-form factuality evaluation on LongFact
90.2AccuracyCURE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CUREPipeline Stage=Factuality Optimization (Full Pipeline)2026.04 | 90.2 | 14.4 | 16.8 | 66.9 | |
| SFT + F-RL + COPipeline Stage=Calibration Optimization2026.04 | 89.8 | 14.6 | 16.9 | 65.6 | |
| SFTPipeline Stage=Supervised Fine-Tuning2026.04 | 85.1 | 11.8 | 14.6 | 65.5 | |
| SFT + F-RLPipeline Stage=Feasibility RL2026.04 | 83.7 | 13.5 | 15.5 | 59.8 | |
| L2RF2026.04 | 79.4 | — | — | — | |
| Base LLMBackbone=Llama3.1-8B-Instruct2026.04 | 74.5 | 15.8 | 19.3 | 59.1 | |
| LitCab2026.04 | 74.5 | 15.6 | 19 | 60.1 |