Global Fact Consistency Verification on FEVER
99.5PrecisionDirect (baseline)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Direct (baseline)Backbone=Claude 42026.01 | 99.5 | 83.3 | 89.1 | |
| Direct (baseline)Backbone=Claude 3.72026.01 | 99.2 | 80.5 | 87.3 | |
| QXRBackbone=GPT-OSS-120B2026.01 | 99.2 | 98 | 98.5 | |
| Direct (baseline)Backbone=DeepSeek-R12026.01 | 98.9 | 82.1 | 87.5 | |
| QXRBackbone=DeepSeek-R12026.01 | 98.8 | 98 | 98.3 | |
| QXRBackbone=Claude 3.72026.01 | 98.3 | 97.7 | 98 | |
| QXRBackbone=Claude 42026.01 | 98.1 | 97.7 | 97.8 | |
| Direct (baseline)Backbone=GPT-OSS-120B2026.01 | 97.6 | 97.5 | 97.6 | |
| Direct (baseline)Backbone=Mistral Large (2407)2026.01 | 97 | 78 | 84.8 | |
| QXRBackbone=Mistral Large (2407)2026.01 | 96.4 | 99 | 97.6 |