BizBench - TAT-QA Extract: leaderboard
Metric: Accuracy (%; 248 TAT-QA questions answerable by extracting a number from the text or table, 3-shot). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 2 70B | 94.4 |
| 2 | Mixtral 8x7B | 91.1 |
| 3 | GPT-4 | 90.3 |
| 4 | Llama 2 13B | 88.7 |
| 5 | Mistral 7B | 87.9 |
| 6 | GPT-3.5 | 84.2 |
| 7 | Llama 2 7B | 83.1 |
| 8 | mpt-30B | 81.5 |
| 9 | falcon-40B | 80.6 |
| 10 | starcoder | 75 |
| 11 | falcon-7B | 62.9 |
| 12 | mpt-7B | 62.1 |
Interactive version: theaggregate.ai/benchmark?slug=bizbench-tat-qa-extract · How It Works · Data refreshed daily, snapshot 2026-09-26.