StatFormBench - Variable Recall: leaderboard
Variable recall on StatFormBench: the share of reference variables for a statistical question that the model identifies (StatFormBench, arXiv 2609.01982v2, Table 1). Percent; higher is better; penalizes missed variables.
Metric: Recall (%). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 72.9 | #67 |
| 2 | Claude Sonnet 4.6 | 70.9 | #94 |
| 3 | DeepSeek V4 Flash | 70.9 | #155 |
| 4 | Kimi K2.5 | 69.8 | #152 |
| 5 | Gemini 3 Flash | 68.9 | #103 |
| 6 | Gemini 3.1 Pro (Preview) | 68.3 | #64 |
| 7 | GPT-5.4 | 68.3 | #83 |
| 8 | DeepSeek V4 Pro | 67.3 | #112 |
| 9 | GPT-5.5 | 67.2 | #33 |
| 10 | Qwen 3.5 397B A17B | 66.5 | #154 |
| 11 | Qwen 3.5 9B | 58.1 | #391 |
| 12 | Qwen 3.5 4B | 54.9 | #514 |
| 13 | Qwen 3.5 2B | 25.2 | #1114 |
| 14 | Qwen 3.5 0.8B | 7.6 | #1410 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=statformbench-variable-recall · How It Works · Data refreshed daily, snapshot 2026-11-05.