Fact-claim Decomposition on WikiSplitBench-1000 1.0 (test)
90.4Semantic F1CREDENCE-base
Evaluation Results
| Method | Links | |
|---|---|---|
| CREDENCE-baseModel=Qwen3-8B2026.06 | 90.4 | |
| CREDENCE-allModel=Qwen3-8B2026.06 | 90.3 | |
| CREDENCE-baseModel=Gemma-3-12b2026.06 | 89.4 | |
| CREDENCE-allModel=Gemma-3-12b2026.06 | 89.4 | |
| CREDENCE-baseModel=Phi-3-mini2026.06 | 89.1 | |
| CREDENCE-allModel=Phi-3-mini2026.06 | 89.1 | |
| FActScore-promptModel=Qwen3-8B2026.06 | 87.1 | |
| CREDENCE-baseModel=Gemini Flash2026.06 | 86.9 | |
| CREDENCE-allModel=Gemini Flash2026.06 | 86.9 | |
| FActScore-promptModel=Gemini Flash2026.06 | 85.3 | |
| FActScore-promptModel=Gemma-3-12b2026.06 | 83.6 | |
| SpaCy-senterModel=Phi-3-mini2026.06 | 80.9 | |
| SpaCy-senterModel=Qwen3-8B2026.06 | 80.9 | |
| SpaCy-senterModel=Gemma-3-12b2026.06 | 80.9 | |
| SpaCy-senterModel=Gemini Flash2026.06 | 80.9 | |
| FActScore-promptModel=Phi-3-mini2026.06 | 31.3 |