StatFormBench - Variable Role Identification: leaderboard
Variable role identification (VRI) on StatFormBench: the fraction of reference variables that a model both recovers and assigns the correct question-conditional role, penalizing missed variables and wrong roles (StatFormBench, arXiv 2609.01982v2, Table 1). Percent; higher is better.
Metric: VRI (%). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 66.2 | #67 |
| 2 | DeepSeek V4 Flash | 65.3 | #155 |
| 3 | Kimi K2.5 | 63.8 | #152 |
| 4 | Gemini 3 Flash | 63.6 | #103 |
| 5 | Claude Sonnet 4.6 | 63.3 | #94 |
| 6 | Gemini 3.1 Pro (Preview) | 63.1 | #64 |
| 7 | GPT-5.4 | 62 | #83 |
| 8 | Qwen 3.5 397B A17B | 61.4 | #154 |
| 9 | GPT-5.5 | 61.1 | #33 |
| 10 | DeepSeek V4 Pro | 60.3 | #112 |
| 11 | Qwen 3.5 9B | 45 | #391 |
| 12 | Qwen 3.5 4B | 41.1 | #514 |
| 13 | Qwen 3.5 2B | 7.1 | #1114 |
| 14 | Qwen 3.5 0.8B | 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-role-identification · How It Works · Data refreshed daily, snapshot 2026-11-05.