StatFormBench - Variable Precision: leaderboard
Variable precision on StatFormBench: the share of variables a model identifies as relevant that match the reference variable set (StatFormBench, arXiv 2609.01982v2, Table 1). Percent; higher is better; penalizes over-prediction.
Metric: Precision (%). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 66.1 | #103 |
| 2 | Claude Opus 4.6 | 65.8 | #67 |
| 3 | Claude Sonnet 4.6 | 65 | #94 |
| 4 | DeepSeek V4 Pro | 64.9 | #112 |
| 5 | Gemini 3.1 Pro (Preview) | 64.5 | #64 |
| 6 | Kimi K2.5 | 61.3 | #152 |
| 7 | Qwen 3.5 397B A17B | 60.5 | #154 |
| 8 | DeepSeek V4 Flash | 56.1 | #155 |
| 9 | GPT-5.4 | 53.7 | #83 |
| 10 | GPT-5.5 | 53.7 | #33 |
| 11 | Qwen 3.5 9B | 51.6 | #391 |
| 12 | Qwen 3.5 4B | 48.7 | #514 |
| 13 | Qwen 3.5 2B | 32 | #1114 |
| 14 | Qwen 3.5 0.8B | 8.3 | #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-precision · How It Works · Data refreshed daily, snapshot 2026-11-05.