TableVista - Structural Noise: leaderboard

Accuracy under the Structural Noise rendering (faded gridlines and geometric warping) of all questions in 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 3,000 TableVista questions under the Structural Noise rendering (faded gridlines and geometric warping), 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-reasoning70.4#138
2Qwen 2.5 VL 72B Instruct54.4#849
3Llama 4 Maverick Instruct FP854.4#1034
4Gemma 4 31B (IT)53.3#354
5Qwen 2.5 VL 32B Instruct49.1#1065
6Qwen 3.5 122B A10BNon-reasoning48.7#350
7GPT-5.4 MiniNon-reasoning48.4#409
8Llama 4 Scout Instruct48.3#1580
9Qwen 3.5 27BNon-reasoning47.3#380
10InternVL3.5-14B45.2#1519
11InternVL3.5-30B-A3B44.9#1269
12Gemma 4 26B A4B (IT)43.4#563
13Qwen 2.5 VL 7B Instruct42.1#1919
14Qwen 3 VL 8B Instruct41.9#1043
15Qwen 3 VL 30B A3B Instruct41.8#903

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=tablevista-structural-noise · How It Works · Data refreshed daily, snapshot 2026-11-22.