Qianfan OCR vs Unlimited OCR
Pick Qianfan OCR for self-hosted multilingual parsing of single pages with tables, formulas and key information extraction, where it scores 93.12 on OmniDocBench v1.5. Pick Unlimited OCR for long reports and books that you want parsed in one request, as it handles 40+ pages in a single pass. Qianfan OCR has the stronger published page-level benchmark and wider output formats, while Unlimited OCR's main advantage is multi-page input.
Qianfan OCRbaidu/qianfan-ocr | Unlimited OCRbaidu/unlimited-ocr | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 2.0 s / page | 1.2 s / page |
| Best for | Chinese documents | Books and very long documents |
| Inputs | PDF, images | PDF, images |
| Languages | Chinese, English and more | 30+ languages |
| Max pages per request | 200 | 1,000 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | ✓ Yes |
| On OCRRouter | Available | Available |
Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.
Which to choose
Choose Qianfan OCR if
- Its 93.12 on OmniDocBench v1.5 is the top end-to-end score in Baidu's table.
- It covers 192 languages, handwriting and scene text.
- It outputs Markdown, JSON, HTML, LaTeX and bounding boxes, which suits tables, formulas and key information extraction.
Choose Unlimited OCR if
- It parses 40+ pages in a single forward pass, with memory staying roughly constant, so long reports and books go in one request.
- Unlimited OCR 2 scores 77.1 on MarathonDocBench against 33.7 for the next model, per Baidu.
- It is MIT licensed with no copyleft, and it returns block types and bounding boxes alongside text for layout-aware extraction.
Key differences
| Qianfan OCR | Unlimited OCR | |
|---|---|---|
| Licence | Apache-2.0, commercial use allowed | MIT, commercial use allowed, no copyleft |
| Model size | 4B, built on the Qwen3-4B language model | 3B Mixture-of-Experts, about 500M active per token (third-party coverage); Unlimited OCR 2 is 3.34B |
| Output formats | Markdown, JSON, HTML, LaTeX, bounding boxes | Markdown, plain text, bounding boxes |
| Multi-page input | Not published | 40+ pages in one pass; context capped at 32,768 tokens; multi-page only works in base mode |
| Hardware | Benchmarked on one NVIDIA A100 with vLLM; no minimum VRAM stated | NVIDIA GPU with CUDA 12.9 or 13.0; Unlimited-OCR-2 under SGLang runs out of memory on 12 GB GPUs |
| Benchmarks | OmniDocBench v1.5 93.12, olmOCR-Bench 79.8, OCRBench 880 (authors) | ParseBench mean 46.17 (independent); MPDocBench 84.6 and MarathonDocBench 77.1 for Unlimited OCR 2 (authors) |
| Known failure modes | Thinking mode can emit up to 16,384 tokens, adding latency; 1.024 pages per second on an A100 even with W8A8 | Text repetition and looping on some pages, reported hallucinated text, ParseBench text formatting score of only 0.97 so bold and italic are mostly lost, bbox coordinate loops on GB10 GPUs under SGLang |
Sources
- Qianfan-OCR model card
- Qianfan-OCR paper
- Hugging Face model card
- MarkTechPost on Unlimited OCR 2
- Issue #55 repetition
- Issue #114 SGLang loops
What each one is
Qianfan OCR
Baidu's OCR-tuned vision model. Strong on Chinese documents and tables.
Qianfan OCR detailsUnlimited OCR
Open-source model from Baidu that reads hundreds of pages in one pass with flat memory use.
Unlimited OCR detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "baidu/qianfan-ocr","fallbacks": ["baidu/unlimited-ocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qianfan OCR alternativesUnlimited OCR alternativesBest OCR models in 2026