HELM TORR - Tablebench Numerical Reasoning: leaderboard
The "Tablebench Numerical Reasoning" subtask of HELM TORR. Stanford HELM's TORR project (Table-Oriented Reasoning and Retrieval): question answering and reasoning over tables and semi-structured records, including financial filings, scored on program and answer accuracy.
Metric: ROUGE-L. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 42.29 | #291 |
| 2 | GPT-4o (2024-11-20) | 39.87 | #378 |
| 3 | DeepSeek V3 | 35.14 | #323 |
| 4 | Gemini 1.5 Pro (002) | 31.48 | #350 |
| 5 | Llama 3.1 405B Instruct | 31.45 | #461 |
| 6 | Llama 3.1 70B Instruct | 26.53 | #566 |
| 7 | GPT-4o Mini (2024-07-18) | 23.55 | #561 |
| 8 | Qwen 2 72B Instruct | 23.44 | #579 |
| 9 | Gemini 1.5 Flash (002) | 21.04 | #585 |
| 10 | Claude 3.5 Haiku (20241022) | 20.33 | #591 |
| 11 | Mixtral 8x22B Instruct | 20.17 | #733 |
| 12 | Mixtral 8x7B Instruct | 12.24 | #1172 |
| 13 | Llama 3.1 8B Instruct | 9.68 | #1040 |
| 14 | Mistral 7B Instruct (v0.3) | 9.65 | #1276 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-torr-tablebench-numerical-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-23.