IndoTabVQA - Arabic: leaderboard

IndoTabVQA (2026) asks questions about tables in real Bahasa Indonesia document images (bordered, borderless and colorful tables). This board uses the Arabic questions, a cross-lingual setting: zero-shot In-Match accuracy, the normalized gold answer found inside the answer of the model, in percent; higher is better.

Metric: In-Match accuracy (%), the normalized gold answer found inside the model's answer, for Arabic questions about tables in real Bahasa Indonesia document images, 1,043 test images, zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 2.5 VL 7B23#1862
2GPT-4o21.4#751
3IndoTabVQA Llama-3.2-11B-Vision (checkpoint unspecified)19.4
4Gemma 3 12B17.4#2059
5Qwen 2.5 VL 3B16.4#2639
6Donut (IndoTabVQA checkpoint unspecified)4.39

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

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