TableVista - Partial: leaderboard

Accuracy under the Partial rendering, which splits each table into disjoint visual blocks, 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 Partial rendering (the table split into disjoint visual blocks), 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-reasoning68.8#138
2Llama 4 Maverick Instruct FP852.9#1034
3Gemma 4 31B (IT)51.2#354
4Qwen 2.5 VL 72B Instruct50.5#849
5Qwen 2.5 VL 32B Instruct48.1#1065
6Qwen 3.5 122B A10BNon-reasoning47.3#350
7GPT-5.4 MiniNon-reasoning46.8#409
8Qwen 3.5 27BNon-reasoning46.7#380
9Llama 4 Scout Instruct46#1580
10InternVL3.5-30B-A3B43.5#1269
11InternVL3.5-14B43.2#1519
12Qwen 2.5 VL 7B Instruct41#1919
13Qwen 3 VL 30B A3B Instruct40.4#903
14Gemma 4 26B A4B (IT)39.9#563
15Qwen 3 VL 8B Instruct39.2#1043

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

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