TableVista - Hard Questions: leaderboard
Metric: Accuracy (%) on the hard questions (most reasoning steps, multi-hop aggregation) 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) | 47 |
| 2 | GPT-5.4 Mini (Non-reasoning) | 35.2 |
| 3 | Llama 4 Maverick Instruct FP8 | 24.4 |
| 4 | Qwen 2.5 VL 72B Instruct | 22.1 |
| 5 | Gemma 4 31B (IT) | 21.9 |
| 6 | Qwen 3.5 27B (Non-reasoning) | 19.6 |
| 7 | Qwen 2.5 VL 32B Instruct | 19.4 |
| 8 | Qwen 3.5 122B A10B (Non-reasoning) | 18.9 |
| 9 | Llama 4 Scout Instruct | 18.8 |
| 10 | Gemma 4 26B A4B (IT) | 16.6 |
| 11 | Qwen 3.6 35B A3B (Non-reasoning) | 16.4 |
| 12 | Qwen 3 VL 30B A3B Instruct | 14.8 |
| 13 | Qwen 2.5 VL 7B Instruct | 14.2 |
| 14 | Gemma 3 27B (IT) | 14.1 |
| 15 | Qwen 3.5 9B (Non-reasoning) | 14.1 |
Interactive version: theaggregate.ai/benchmark?slug=tablevista-hard-questions · How It Works · Data refreshed daily, snapshot 2026-10-07.