MMTU: leaderboard
Massive Multi-Task Table Understanding benchmark with real-world table reasoning tasks including joins, transformations, cleaning, and question answering.
Metric: Accuracy (%). Source: huggingface.co. Source: huggingface.co. Saturation forecast: Around January 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 69.58 | #98 |
| 2 | O3 | 69.13 | #124 |
| 3 | GPT-5 Mini | 66.74 | #186 |
| 4 | Gemini 2.5 Pro | 66.45 | #152 |
| 5 | Grok 3 Mini | 64.55 | #281 |
| 6 | Gemini 2.5 Flash | 62.55 | #245 |
| 7 | DeepSeek R1 0528 | 57.99 | #222 |
| 8 | GPT-5 Chat | 57.67 | #241 |
| 9 | GPT-OSS-120B | 54.32 | #341 |
| 10 | Qwen 3 235B A22B 2507 (Thinking) | 52.87 | #260 (Qwen 3 235B A22B 2507) |
| 11 | Qwen 3 235B A22B 2507 Instruct | 52.42 | #300 |
| 12 | GPT-4o (2024-11-20) | 50.73 | #381 |
| 13 | Qwen 3 32B | 50.6 | #443 |
| 14 | Llama 4 Maverick Instruct FP8 | 49 | #414 |
| 15 | GPT-OSS-20B | 47.8 | #518 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmtu · How It Works · Data refreshed daily, snapshot 2026-10-20.