IDP Leaderboard: leaderboard
Nanonets and IIT Indore document-AI leaderboard (2025) scoring models on OCR, table extraction, structured field extraction (KIE) and visual QA over document images; edit-distance accuracy and GriTS.
Metric: Avg Score (%). Source: idp-leaderboard.org. Saturation forecast: Around January 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 81 | #79 |
| 2 | GPT-5.2 | 79.1 | #110 |
| 3 | Gemini 3 Pro | 77.7 | #80 |
| 4 | Qwen 3.5 9B | 77.2 | #379 |
| 5 | Qwen 3 VL 235B A22B | 76.8 | #238 |
| 6 | Qwen 3.5 4B | 75.4 | #475 |
| 7 | Gemini 3 Flash | 75.3 | #99 |
| 8 | Claude Opus 4.6 | 74.1 | #63 |
| 9 | Claude Sonnet 4.6 | 73.9 | #90 |
| 10 | Qwen 3.5 2B | 71.9 | #1148 |
| 11 | GPT-5 Mini | 69.9 | #186 |
| 12 | Gemini 3.1 Pro (Preview) | 69.8 | #58 |
| 13 | Mistral Small 4 | 69.6 | #517 |
| 14 | Qwen 3.5 0.8B | 64.8 | #1424 |
| 15 | Claude Haiku 4.5 | 61.2 | #280 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=idp-leaderboard · How It Works · Data refreshed daily, snapshot 2026-10-20.