Chain of Physics FailureSensorIQ - CoP: leaderboard
Chain of Physics FailureSensorIQ - CoP: Diagnostic accuracy on 1314 single-correct questions (%). Chain of Physics arXiv 2611.14263v1: diagnostic accuracy on 1314 single-correct FailureSensorIQ questions. Direct, RAG, full CoP and ablated CoP remain separate evaluation runs. One run/model, temperature zero. Claude Opus 4.7 is retained. The paper never identifies whether bare Llama 3.1 8B and Qwen 2.5 7B mean base or instruct, so these rows are held rather than silently crossing that boundary.
Metric: Diagnostic accuracy on 1314 single-correct questions (%). Source: arxiv.org. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 (Benchmark Sweep Run: source=chain_physics_paper; harness=CoP; temperature=0; effort=unreported) | 68.8 |
Interactive version: theaggregate.ai/benchmark?slug=chain-of-physics-failuresensoriq-cop · How It Works · Data refreshed daily, snapshot 2026-11-12.