TableVision: leaderboard
TableVision (2026) asks vision-language models questions about images of complex tables with multi-level hierarchical headers, from cell retrieval to multi-hop and temporal analysis, over a 1,301-question test set. Exact-match accuracy (%), zero-shot; higher is better.
Metric: Exact-match accuracy (%) over all 1,301 test questions, zero-shot, the model answers directly from the table image; TableVision test set; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Glyph | 71.6 | |
| 2 | Qwen 3 VL 8B Instruct | 59.3 | #1029 |
| 3 | Ovis2.5-9B | 58.9 | #1225 |
| 4 | GLM-4.1V-9B | 51 | #1125 |
| 5 | MiniCPM-V-2.6 | 48.1 | #3267 |
| 6 | HiPPO | 46.8 | |
| 7 | DeepSeek-VL2 | 39 | #2927 |
| 8 | LLaVA-OneVision-7B | 34.6 | #2981 |
| 9 | InternVL3-8B | 28.8 | #1782 |
| 10 | Phi-3.5-Vision Instruct | 19.9 | #6268 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=tablevision · How It Works · Data refreshed daily, snapshot 2026-11-21.