C2-Faith - Causal Step Localization: leaderboard
Metric: Exact-match localization rate (%): the share of perturbed chains for which the judge returns the exact index of the injected acausal step (output -1 counts as a miss), on 450 perfect PRM800K MATH solution chains with one middle step replaced by an LLM-written logically inconsistent rewrite, default API parameters; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O4 Mini | 68 | #172 |
| 2 | GPT-4.1 | 57.6 | #240 |
| 3 | DeepSeek V3.1 | 55.8 | #260 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=c2-faith-causal-step-localization · How It Works · Data refreshed daily, snapshot 2026-10-11.