MMTU - KB mapping: leaderboard
The "KB mapping" 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 January 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 | 61.44 | #124 |
| 2 | GPT-5 | 61.18 | #98 |
| 3 | Gemini 2.5 Pro | 60.84 | #152 |
| 4 | Grok 3 Mini | 54.8 | #281 |
| 5 | GPT-5 Chat | 54.27 | #241 |
| 6 | GPT-5 Mini | 52.77 | #186 |
| 7 | GPT-4o (2024-11-20) | 49.84 | #381 |
| 8 | DeepSeek R1 0528 | 49 | #222 |
| 9 | Qwen 3 235B A22B 2507 (Thinking) | 47.63 | #260 (Qwen 3 235B A22B 2507) |
| 10 | Gemini 2.5 Flash | 47.16 | #245 |
| 11 | Mistral Large 2 (Nov) Instruct (2411) | 45.67 | #465 |
| 12 | Llama 3.3 70B | 43.55 | #579 |
| 13 | GPT-OSS-120B | 42.32 | #341 |
| 14 | Llama 4 Maverick Instruct FP8 | 42.02 | #414 |
| 15 | Qwen 3 235B A22B 2507 Instruct | 41.29 | #300 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmtu-kb-mapping · How It Works · Data refreshed daily, snapshot 2026-10-20.