Massive Multi-Task Table Understanding benchmark with 28,136 questions across 25 task categories testing ability to comprehend, reason about, and extract information from tables in document images.
Largest table understanding benchmark; tests 25 distinct table tasks; highly unsaturated (GPT-5 at 69.6%); directly measures enterprise table capability.
Newer benchmark with fewer models tested; task categories have uneven difficulty.
Higher is better. Overall accuracy across 25 table understanding tasks, 0-100%.
Max score: 100
Newer benchmark with fewer models tested.
| # | Model | Score |
|---|---|---|
| 1 | 75.1 | |
| 2 | 73.9 | |
| 3 | 73.7 | |
| 4 | 73.5 | |
| 5 | 73.3 | |
| 6 | 73.2 | |
| 7 | 73.2 | |
| 8 | 72.7 | |
| 9 | 71.7 | |
| 10 | 71.7 | |
| 11 | 71.3 | |
| 12 | 71.1 | |
| 13 | 70.6 | |
| 14 | 70.5 | |
| 15 | 70.4 | |
| 16 | 70.2 | |
| 17 | 70.1 | |
| 18 | 70.0 | |
| 19 | 69.9 | |
| 20 | 69.6 | |
| 21 | 69.6 | |
| 22 | 69.5 | |
| 23 | 69.5 | |
| 24 | 69.4 | |
| 25 | 69.3 | |
| 26 | 69.3 | |
| 27 | 69.2 | |
| 28 | 69.1 | |
| 29 | 69.1 | |
| 30 | 69.0 | |
| 31 | 68.8 | |
| 32 | 68.8 | |
| 33 | 68.8 | |
| 34 | 68.7 | |
| 35 | 68.7 | |
| 36 | 68.6 | |
| 37 | 68.3 | |
| 38 | 68.2 | |
| 39 | 68.1 | |
| 40 | 68.1 | |
| 41 | 68.1 | |
| 42 | 68.0 | |
| 43 | 68.0 | |
| 44 | 67.9 | |
| 45 | 67.9 | |
| 46 | 67.8 | |
| 47 | 67.8 | |
| 48 | 67.8 | |
| 49 | 67.6 | |
| 50 | 67.3 |