TableVista - Complex Structure: leaderboard

Accuracy on the complex tables with multi-level headers and merged cells (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 1,000 complex tables (multi-level headers, merged cells) 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-reasoning81.7#135
2Qwen 2.5 VL 72B Instruct59.7#841
3GPT-5.4 MiniNon-reasoning59.3#406
4Gemma 4 31B (IT)57.6#350
5Llama 4 Maverick Instruct FP855.9#1023
6Qwen 2.5 VL 32B Instruct54.5#1062
7Qwen 3.5 27BNon-reasoning53#375
8Qwen 3.5 122B A10BNon-reasoning52.2#346
9Gemma 4 26B A4B (IT)50.1#556
10Llama 4 Scout Instruct49.6#1579
11InternVL3.5-14B48.9#1504
12InternVL3.5-30B-A3B47.7#1267
13Qwen 2.5 VL 7B Instruct47.5#1902
14Qwen 3 VL 30B A3B Instruct46.8#895
15Molmo2-8B45.2#1621

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

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