MMTU - Data Cleaning: leaderboard
The "Data Cleaning" 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: Around February 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 54.57 | #152 |
| 2 | O3 | 51.75 | #124 |
| 3 | GPT-5 | 51.05 | #98 |
| 4 | GPT-5 Mini | 49.22 | #186 |
| 5 | GPT-5 Chat | 45.64 | #241 |
| 6 | GPT-4o (2024-11-20) | 44.55 | #381 |
| 7 | Gemini 2.5 Flash | 43.7 | #245 |
| 8 | Qwen 3 235B A22B 2507 Instruct | 42.6 | #300 |
| 9 | Grok 3 Mini | 41.93 | #281 |
| 10 | Llama 4 Maverick Instruct FP8 | 40.2 | #414 |
| 11 | DeepSeek R1 0528 | 39.93 | #222 |
| 12 | Llama 3.3 70B | 38.64 | #579 |
| 13 | Mistral Large 2 (Nov) Instruct (2411) | 37.31 | #465 |
| 14 | Qwen 3 32B | 33.8 | #443 |
| 15 | Qwen 3 8B | 32.63 | #674 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmtu-data-cleaning · How It Works · Data refreshed daily, snapshot 2026-10-20.