TableVista - Screen Capture: leaderboard
Metric: Accuracy (%) on the 3,000 TableVista questions under the vision-only Screen Capture setting (question, context and table rendered in one image), 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 January 2027. 29 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (Non-reasoning) | 69.4 |
| 2 | Qwen 2.5 VL 72B Instruct | 57.4 |
| 3 | Qwen 3.5 27B (Non-reasoning) | 55.1 |
| 4 | Qwen 3.5 122B A10B (Non-reasoning) | 54.3 |
| 5 | Llama 4 Maverick Instruct FP8 | 53.4 |
| 6 | Qwen 2.5 VL 32B Instruct | 51.1 |
| 7 | Llama 4 Scout Instruct | 46.7 |
| 8 | Qwen 3.6 35B A3B (Non-reasoning) | 46.1 |
| 9 | Qwen 3 VL 8B Instruct | 45.3 |
| 10 | Gemma 4 31B (IT) | 42.2 |
| 11 | GPT-5.4 Mini (Non-reasoning) | 42 |
| 12 | Qwen 2.5 VL 7B Instruct | 41.6 |
| 13 | Qwen 3 VL 30B A3B Instruct | 41.1 |
| 14 | Gemma 3 27B (IT) | 40.1 |
| 15 | Qwen 3.5 9B (Non-reasoning) | 39.7 |
Interactive version: theaggregate.ai/benchmark?slug=tablevista-screen-capture · How It Works · Data refreshed daily, snapshot 2026-10-07.