GRACE - Error Categorization: leaderboard
Metric: Error-categorization macro-F1 (%) averaged over the two tracks: identifying the specific four-class faithfulness-error category of each unfaithful step; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 72.08 |
| 2 | Gemini 3 Flash | 63.36 |
| 3 | Gemma 4 31B | 58.5 |
| 4 | GPT-5.4 | 58.43 |
| 5 | Qwen 3.5 27B | 54.36 |
| 6 | Qwen 3.5 35B A3B | 47.15 |
| 7 | GPT-5.4 Mini | 41.57 |
| 8 | Qwen 3 14B | 35.17 |
| 9 | Qwen 3 8B | 27.21 |
| 10 | Llama 3.1 8B | 26.88 |
Interactive version: theaggregate.ai/benchmark?slug=grace-error-categorization · How It Works · Data refreshed daily, snapshot 2026-09-29.