StatFormBench - Fine-Grained Classification: leaderboard
Accuracy at identifying the implied statistical problem type among 85 fine-grained categories on 1,013 StatFormBench samples; a correct fine-grained label implies a correct coarse-grained one (StatFormBench, arXiv 2609.01982v2, Table 1). Percent; higher is better.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 72 | #64 |
| 2 | Claude Opus 4.6 | 69.5 | #67 |
| 3 | Gemini 3 Flash | 68.4 | #103 |
| 4 | GPT-5.5 | 67.2 | #33 |
| 5 | GPT-5.4 | 65.8 | #83 |
| 6 | Kimi K2.5 | 64.8 | #152 |
| 7 | DeepSeek V4 Flash | 64.8 | #155 |
| 8 | Qwen 3.5 397B A17B | 64.6 | #154 |
| 9 | Claude Sonnet 4.6 | 63.9 | #94 |
| 10 | DeepSeek V4 Pro | 59.1 | #112 |
| 11 | Qwen 3.5 4B | 30.6 | #514 |
| 12 | Qwen 3.5 9B | 20.3 | #391 |
| 13 | Qwen 3.5 0.8B | 3.8 | #1410 |
| 14 | Qwen 3.5 2B | 3.5 | #1114 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=statformbench-fine-grained-classification · How It Works · Data refreshed daily, snapshot 2026-11-05.