HELM TORR - Wikitq: leaderboard
The "Wikitq" 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: F1 Strings. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 91.94 | #291 |
| 2 | DeepSeek V3 | 90.8 | #323 |
| 3 | GPT-4o (2024-11-20) | 90.75 | #378 |
| 4 | Llama 3.1 70B Instruct | 90.72 | #566 |
| 5 | Llama 3.1 405B Instruct | 90.2 | #461 |
| 6 | Gemini 1.5 Pro (002) | 88.53 | #350 |
| 7 | GPT-4o Mini (2024-07-18) | 88.42 | #561 |
| 8 | Gemini 1.5 Flash (002) | 88.07 | #585 |
| 9 | Qwen 2 72B Instruct | 86.2 | #579 |
| 10 | Claude 3.5 Haiku (20241022) | 84.98 | #591 |
| 11 | Llama 3.1 8B Instruct | 80.01 | #1040 |
| 12 | Mistral 7B Instruct (v0.3) | 67.57 | #1276 |
| 13 | Mixtral 8x22B Instruct | 66.78 | #733 |
| 14 | Mixtral 8x7B Instruct | 61.91 | #1172 |
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-wikitq · How It Works · Data refreshed daily, snapshot 2026-10-23.