GRACE - ReClor Categorization: leaderboard
Metric: Four-class error-categorization macro-F1 (%) on ReClor inference traces; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 67.44 |
| 2 | Gemma 4 31B | 64.13 |
| 3 | Gemini 3 Flash | 62.67 |
| 4 | GPT-5.4 | 55.2 |
| 5 | GPT-5.4 Mini | 45.93 |
| 6 | Qwen 3.5 27B | 44.39 |
| 7 | Qwen 3.5 35B A3B | 39.4 |
| 8 | Qwen 3 14B | 29.91 |
| 9 | Llama 3.1 8B | 26.19 |
| 10 | Qwen 3 8B | 22.15 |
Interactive version: theaggregate.ai/benchmark?slug=grace-reclor-categorization · How It Works · Data refreshed daily, snapshot 2026-09-29.