LLM Stats (DocVQA): leaderboard
Visual QA on scanned document images (forms, reports, letters), requiring extraction of information from handwritten or printed text within document layouts.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Estimated already saturated. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 72B Instruct | 96.4 | #377 |
| 2 | Qwen 2.5 VL 7B Instruct | 95.7 | #654 |
| 3 | Claude 3.5 Sonnet | 95.2 | #350 |
| 4 | Qwen2.5-Omni-7B | 95.2 | #624 |
| 5 | Mistral Small 3.2 | 94.86 | #605 |
| 6 | Qwen 2.5 VL 32B Instruct | 94.8 | #453 |
| 7 | Llama 4 Maverick | 94.4 | #464 |
| 8 | Llama 4 Scout | 94.4 | #656 |
| 9 | Grok 2 | 93.6 | #452 |
| 10 | Nova Pro | 93.5 | #569 |
| 11 | Phi-4 Multimodal Instruct | 93.2 | #906 |
| 12 | GPT-4o | 92.8 | #343 |
| 13 | Nova Lite | 92.4 | #754 |
| 14 | Pixtral-12B | 90.7 | #808 |
| 15 | Llama 3.2 90B Instruct | 90.1 | #601 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-docvqa · How It Works · Data refreshed daily, snapshot 2026-10-20.