Faithfulness detection on In-domain Step-level Benchmark Agent
80.2FF1GeoFaith
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GeoFaith2026.05 | 80.2 | 69.8 | |
| GPT-o12026.05 | 77.1 | 66.8 | |
| GPT-4o2026.05 | 76.2 | 56 | |
| DeepSeek-V32026.05 | 72.3 | 57.9 | |
| o3-mini2026.05 | 72.1 | 53.5 | |
| Llama-3.1-70B-Instruct2026.05 | 71.8 | 49.7 | |
| Qwen2.5-32B-Instruct2026.05 | 70.9 | 62.3 | |
| LogicReward2026.05 | 69.7 | 51 | |
| FaithLens2026.05 | 53.3 | 36.5 | |
| HHEM2.12026.05 | 49.7 | 32.3 |