DocVQA 2026: leaderboard
DocVQA 2026 document reasoning accuracy across eight domains; the published baseline uses high thinking and counts oversized API inputs as failures.
Metric: Accuracy (%, high thinking). Source: huggingface.co. Saturation forecast: Around March 2028. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) (High) | 37.5 | #71 (Gemini 3 Pro (Preview)) |
| 2 | GPT-5.2 (High) | 35 | #115 (GPT-5.2) |
| 3 | Gemini 3 Flash (Preview) (High) | 33.75 | #90 (Gemini 3 Flash (Preview)) |
| 4 | GPT-5 Mini (High) | 22.5 | #193 (GPT-5 Mini) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=docvqa-2026 · How It Works · Data refreshed daily, snapshot 2026-11-05.