GRACE - ReClor Categorization: leaderboard
GRACE is a step-level faithfulness benchmark for context-grounded chain-of-thought reasoning, with human-annotated traces from four datasets on two tracks (grounding: MuSiQue, 2WikiMHQA; inference: LogiQA, ReClor) and a data-driven error taxonomy. This board reports Four-class error-categorization macro-F1 (%) on ReClor inference traces; higher is better.
Metric: Four-class error-categorization macro-F1 (%) on ReClor inference traces; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 67.44 | #94 |
| 2 | Gemma 4 31B | 64.13 | #394 |
| 3 | Gemini 3 Flash | 62.67 | #164 |
| 4 | GPT-5.4 | 55.2 | #123 |
| 5 | GPT-5.4 Mini | 45.93 | #389 |
| 6 | Qwen 3.5 27B | 44.39 | #359 |
| 7 | Qwen 3.5 35B A3B | 39.4 | #482 |
| 8 | Qwen 3 14B | 29.91 | #1233 |
| 9 | Llama 3.1 8B | 26.19 | #4861 |
| 10 | Qwen 3 8B | 22.15 | #1830 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=grace-reclor-categorization · How It Works · Data refreshed daily, snapshot 2026-11-13.