idp-leaderboard.org is built with KaTeX, hosted on AWS.
| Technology | Version | First seen |
|---|---|---|
| JavaScript Frameworks & Libraries3 | ||
| – | 9 Nov 2026 | |
| 16.1.6 | 9 Nov 2026 | |
| Frontend Build Tools & CSS3 | ||
| – | 9 Nov 2026 | |
| Web Servers, Languages & Backend Frameworks1 | ||
| CDN & Hosting2 | ||
| – | 9 Nov 2026 | |
| Fonts & Icon Fonts2 | ||
| – | 9 Nov 2026 | |
Registration from the registry's RDAP record, the certificate the site serves, and its rank in public traffic lists.
The crew hasn’t read this site’s colors yet. We read a palette when we capture a site’s screenshot, busiest sites first. Meanwhile, browse palettes from other sites.
Counted from every stack change we've recorded: of the sites we watched running one of this site's tools, how many went on to add another. A pattern across sites, not a forecast for this one.
ns-300.awsdns-37.comns-1612.awsdns-09.co.ukns-1443.awsdns-52.orgns-932.awsdns-52.netgoogle-site-verification=dqc0kOKATFe_GGRxi8Xoz5AdxMLkXK7EvUPOIJhf3gYUser-Agent: * Allow: / User-Agent: GPTBot Allow: / User-Agent: ChatGPT-User Allow: / User-Agent: Google-Extended Allow: / User-Agent: Anthropic-AI Allow: / User-Agent: ClaudeBot Allow: / User-Agent: PerplexityBot Allow: / Sitemap: https://idp-leaderboard.nanonets.com/sitemap.xml
# IDP Leaderboard > Comprehensive benchmark comparing document AI models across OCR, table extraction, key information extraction, and visual question answering. Built by Nanonets. The Intelligent Document Processing (IDP) Leaderboard evaluates AI models on three complementary benchmarks: 1. **OlmOCR Bench** — OCR quality on math equations, tables, headers/footers, multi-column reading order (~7,000 document pages) 2. **OmniDocBench v1.5** — Document parsing: text extraction (edit distance), formula recognition (CDM), table structure (TEDS), reading order 3. **IDP Core Bench** — Real-world tasks: key information extraction (KIE), OCR on handwriting, table understanding (GrITS), visual question answering (VQA) Overall score = unweighted mean of three benchmark scores. ## Current Rankings (March 2026) | Rank | Model | Company | Overall | OlmOCR | OmniDoc | IDP | |------|-------|---------|---------|--------|---------|-----| | 1 | Gemini 3.1 Pro | Google | 83.2 | 74.6 | 85.3 | 89.6 | | 2 | Nanonets OCR2+ | Nanonets | 81.8 | 82.2 | 89.5 | 73.8 | | 3 | Gemini-3-Pro | Google | 81.4 | 73.5 | 88.8 | 81.8 | | 4 | GPT-5.4 | OpenAI | 81.0 | 73.4 | 85.2 | 84.4 | | 5 | Claude Sonnet 4.6 | Anthropic | 80.8 | 74.4 | 86.9 | 81.2 | | 6 | Claude Opus 4.6 | Anthropic | 80.3 | 73.9 | 85.9 | 81.1 | | 7 | Gemini-3-Flash | Google | 79.9 | 69.2 | 90.1 | 80.5 | | 8 | GPT-5.2 | OpenAI | 79.2 | 72.2 | 88.0 | 77.4 | | 9 | Qwen3.5-9B | Alibaba | 77.0 | 78.1 | 76.7 | 76.2 | | 10 | Qwen3.5-4B | Alibaba | 73.1 | 77.2 | 67.6 | 74.5 | | 11 | GPT-5-Mini | OpenAI | 70.8 | 56.7 | 82.5 | 73.3 | | 12 | GPT-4.1 | OpenAI | 70.0 | 55.5 | 79.9 | 74.7 | | 13 | Claude Haiku 4.5 | Anthropic | 69.6 | 56.2 | 79.6 | 72.9 | | 14 | Ministral-8B | Mistral AI | 69.2 | 57.8 | 78.2 | 71.7 | | 15 | GLM-OCR | Zhipu AI | 63.6 | 66.7 | 69.2 | 54.9 | | 16 | Qwen3.5-2B | Alibaba | 63.2 | 73.7 | 48.7 | 67.1 | | 17 | Qwen3.5-0.8B | Alibaba | 58.0 | 65.6 | 47.3 | 61.2 | | 18 | GPT-5-Nano | OpenAI | 50.7 | 22.8 | 63.4 | 65.8 | | 19 | Llama-3.2-Vision-11B | Meta | 50.1 | 47.2 | 44.6 | 58.6 | | 20 | Pixtral-12B | Mistral AI | 46.0 | 36.8 | 42.3 | 59.0 | Additionally, Datalab Marker (82.1 OlmOCR, 85.5 OmniDoc) is evaluated on extract-only benchmarks (no IDP/VQA). ## Key Findings - **Best overall**: Gemini 3.1 Pro (83.2%) leads, followed by Nanonets OCR2+ (81.8%) - **Best OCR**: Nanonets OCR2+ (82.2%) leads OlmOCR with strongest table and reading order scores - **Best document parsing**: Gemini-3-Flash (90.1%) leads OmniDocBench with lowest text edit distance - **Best IDP tasks**: Gemini 3.1 Pro (89.6%) leads real-world document processing - **Best open-source**: Qwen3.5-9B (77.0%) is the best open model overall, outperforming all closed models below rank 8 - **Qwen3.5 scaling**: Strong scaling from 0.8B (58.0%) → 2B (63.2%) → 4B (73.1%) → 9B (77.0%), with competitive OlmOCR scores even at small sizes - **Anthropic strong showing**: Claude Sonnet 4.6 (80.8%) and Claude Opus 4.6 (80.3%) in top 6 ## Features - **Leaderboard**: Interactive table and graph views with benchmark and model type filters - **1v1 Compare**: Side-by-side radar chart comparison of any two models - **Results Explorer**: Browse per-sample predictions, ground truth, images, and scores for every model across all benchmarks - **Model Profiles**: Detailed benchmark breakdowns, radar charts, and metadata per model ## Pages - [Home / Leaderboard](https://idp-leaderboard.nanonets.com): Interactive leaderboard with filters - [Benchmarks](https://idp-leaderboard.nanonets.com/benchmarks): OlmOCR, OmniDocBench, IDP Core details - [Models](https://idp-leaderboard.nanonets.com/models): Browse all evaluated models - [1v1 Compare](https://idp-leaderboard.nanonets.com/compare): Side-by-side model comparison - [Results Explorer](https://idp-leaderboard.nanonets.com/explore): Per-sample predictions and images - [GitHub](https://github.com/NanoNets/docext): Source code and evaluation methodology ## Optional - [llms-full.txt](https://idp-leaderboard.nanonets.com/llms-full.txt): Full model data with all metrics
| – |
| 9 Nov 2026 |
| – |
| 9 Nov 2026 |
| – | 9 Nov 2026 |
| – |
| 9 Nov 2026 |
| – |
| 9 Nov 2026 |
| – |
| 9 Nov 2026 |