TableVista - Multi-Table: leaderboard
Metric: Accuracy (%) on the 700 multi-table questions of the 3,000 TableVista questions, Web rendering, direct-output prompt with thinking disabled (reasoning effort none for the GPT models), exact match with a GPT-5-mini semantic-equivalence check on exact-match failures; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 29 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (Non-reasoning) | 61.3 |
| 2 | Qwen 2.5 VL 72B Instruct | 53.1 |
| 3 | Llama 4 Maverick Instruct FP8 | 52.4 |
| 4 | Gemma 4 31B (IT) | 52.3 |
| 5 | Qwen 2.5 VL 32B Instruct | 49.1 |
| 6 | Qwen 3.5 27B (Non-reasoning) | 48.6 |
| 7 | Llama 4 Scout Instruct | 48.1 |
| 8 | Qwen 3.5 122B A10B (Non-reasoning) | 46.3 |
| 9 | Qwen 2.5 VL 7B Instruct | 42.4 |
| 10 | Gemma 4 26B A4B (IT) | 42 |
| 11 | Qwen 3 VL 30B A3B Instruct | 41 |
| 12 | Qwen 3.6 35B A3B (Non-reasoning) | 41 |
| 13 | GPT-5.4 Mini (Non-reasoning) | 40 |
| 14 | Qwen 3 VL 8B Instruct | 39.9 |
| 15 | Gemma 3 27B (IT) | 39.6 |
Interactive version: theaggregate.ai/benchmark?slug=tablevista-multi-table · How It Works · Data refreshed daily, snapshot 2026-10-07.