TableVista - Hard Questions: leaderboard

Accuracy on the hard questions, which need multi-hop aggregation (Web-style rendering), of TableVista (2026), a multimodal table reasoning benchmark in which models answer questions about rendered table images directly with thinking disabled, scored by exact match backed by a semantic-equivalence check. Percent; higher is better.

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 February 2027. 29 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.4Non-reasoning47#135
2GPT-5.4 MiniNon-reasoning35.2#406
3Llama 4 Maverick Instruct FP824.4#1023
4Qwen 2.5 VL 72B Instruct22.1#841
5Gemma 4 31B (IT)21.9#350
6Qwen 3.5 27BNon-reasoning19.6#375
7Qwen 2.5 VL 32B Instruct19.4#1062
8Qwen 3.5 122B A10BNon-reasoning18.9#346
9Llama 4 Scout Instruct18.8#1579
10Gemma 4 26B A4B (IT)16.6#556
11Qwen 3.6 35B A3BNon-reasoning16.4#481
12Qwen 3 VL 30B A3B Instruct14.8#895
13Qwen 2.5 VL 7B Instruct14.2#1902
14Gemma 3 27B (IT)14.1#1324
15Qwen 3.5 9BNon-reasoning14.1#828

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=tablevista-hard-questions · How It Works · Data refreshed daily, snapshot 2026-11-21.