HELM TORR - Tab Fact - Robustness: leaderboard
HELM ToRR (release v1.0.0) robustness on TabFact (verifying statements against tables; 41 instances): per instance, 1 minus the spread between the best and worst Accuracy across 35 prompt configurations (7 table serializations x original, row-shuffled, column-shuffled, transposed and padded tables), averaged. 0-100, higher is more consistent; a model that fails every configuration also scores high.
Metric: Robustness (0-100). Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 67 | #302 |
| 2 | GPT-4o (2024-11-20) | 57 | #392 |
| 3 | Gemini 1.5 Pro (002) | 55 | #368 |
| 4 | DeepSeek V3 | 50 | #329 |
| 5 | Llama 3.1 405B Instruct | 47 | #470 |
| 6 | Mixtral 8x22B Instruct | 47 | #738 |
| 7 | Gemini 1.5 Flash (002) | 43 | #587 |
| 8 | Qwen 2 72B Instruct | 40 | #579 |
| 9 | Claude 3.5 Haiku (20241022) | 36 | #590 |
| 10 | Llama 3.1 70B Instruct | 26 | #567 |
| 11 | GPT-4o Mini (2024-07-18) | 26 | #568 |
| 12 | Mistral 7B Instruct (v0.3) | 21 | #1255 |
| 13 | Mixtral 8x7B Instruct | 19 | #1085 |
| 14 | Llama 3.1 8B Instruct | 2 | #1017 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-torr-tab-fact-robustness · How It Works · Data refreshed daily, snapshot 2026-11-05.