ClaimBench - Subtle contradiction
Professional Work · 2026-10-20
ClaimBench claims made by a small, non-numeric, non-negation edit to a sentence of the excerpt (gold label contradicted). Diagnostic split.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Haiku 5.5 | 100.0 |
| Claude Opus 5.5 | 100.0 |
| Claude Sonnet 5.5 | 100.0 |
| GPT-6.1 Sol | 100.0 |
| GPT-6 Luna | 95.9 |