MMTU - Table Understanding: leaderboard
The "Table Understanding" subtask of MMTU. Massive Multi-Task Table Understanding benchmark with real-world table reasoning tasks including joins, transformations, cleaning, and question answering.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 93.25 | #186 |
| 2 | GPT-5 | 92.95 | #98 |
| 3 | O3 | 88.34 | #124 |
| 4 | Gemini 2.5 Pro | 84.24 | #152 |
| 5 | Gemini 2.5 Flash | 82.04 | #245 |
| 6 | Grok 3 Mini | 78.99 | #281 |
| 7 | GPT-OSS-120B | 74.73 | #341 |
| 8 | GPT-OSS-20B | 68.48 | #518 |
| 9 | Qwen 3 235B A22B 2507 (Thinking) | 67.34 | #260 (Qwen 3 235B A22B 2507) |
| 10 | GPT-5 Chat | 59.13 | #241 |
| 11 | Qwen 3 235B A22B 2507 Instruct | 57.62 | #300 |
| 12 | DeepSeek R1 0528 | 53.81 | #222 |
| 13 | GPT-4o (2024-11-20) | 46.97 | #381 |
| 14 | Llama 4 Maverick Instruct FP8 | 42.12 | #414 |
| 15 | Qwen 3 32B | 26.46 | #443 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmtu-table-understanding · How It Works · Data refreshed daily, snapshot 2026-10-20.