PaddleOCR (PP-OCR) vs Chandra
Pick PaddleOCR (PP-OCR) when you need fast, free text-line and box extraction from printed Chinese, English or other Latin-script text, on CPU, edge devices or in the browser. Pick Chandra when you need whole pages turned into Markdown, HTML or JSON with tables, forms and math, and you can run a GPU or pay for the hosted API. The main split is scope: PP-OCR returns text lines and boxes only, while Chandra is a page-level document model.
PaddleOCR (PP-OCR)paddle/paddleocr | Chandradatalab/chandra | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 0.7 s / page | 2.0 s / page |
| Best for | Fast printed text, Chinese and English | Complex layouts and handwriting |
| Inputs | PDF, images | PDF, images |
| Languages | 80+ languages | 40+ languages |
| Max pages per request | 500 | 500 |
| Tables | No | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | No | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | ✓ Yes | No |
| Forms | No | ✓ Yes |
| 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 PaddleOCR (PP-OCR) if
- It is Apache-2.0 and free to self-host, with no revenue cap on commercial use.
- The PP-OCRv6 medium model is 34.5M parameters and runs on CPU, NVIDIA GPU, Kunlunxin XPU and other accelerators, with a 5.2x CPU speedup using OpenVINO.
- One PP-OCRv6 model covers 50 languages, including Chinese, English, Japanese and 46 Latin-script languages, and the PaddleOCR.js SDK runs it in the browser.
Choose Chandra if
- It converts pages to Markdown, HTML or JSON with layout chunks and bounding boxes per block, so tables, forms and math come out structured.
- Chandra 2 scores 85.8 on olmOCR-bench in the authors' table, above dots.ocr 1.5 (83.9) and olmOCR 2 (82.4), and averages 77.8% across 43 languages.
- You can self-host it with vLLM or call the Datalab API at $1, $3 or $7.50 per 1,000 pages, with a free tier of $20 per month.
Key differences
| PaddleOCR (PP-OCR) | Chandra | |
|---|---|---|
| What it returns | Text lines, bounding boxes and JSON. Tables and formulas need the separate PP-Structure or VL pipelines. | Markdown, HTML, JSON and bounding boxes, with layout labels per block. |
| Licence | Apache-2.0. | Code Apache-2.0. Weights under modified OpenRAIL-M: free for research, personal use and companies under $2M funding or revenue, commercial licence above that. |
| How you run it | Library: paddleocr Python package, C++ or the PaddleOCR.js browser SDK. | Open weights via vLLM or Hugging Face, or the hosted Datalab API. |
| Price | Free, self-hosted. You pay for compute. | Self-hosted free within licence terms. Hosted API $1 (fast), $3 (balanced) or $7.50 (accurate) per 1,000 pages. |
| Hardware | CPU, NVIDIA GPU, Kunlunxin XPU and other accelerators. PP-OCRv6 reports 0.13s on an A100. | Single NVIDIA GPU via vLLM, with presets for A100, L40S and RTX 4090/5090. About 1.44 pages/s on one H100. |
| Benchmarks | Not published | olmOCR-bench overall 85.8 (Chandra 2, authors). 77.8% average on a 43-language set. 72.7% on a 90-language set. |
| Known failure | PP-OCRv6 weights contain denormals that make CPU inference 4 to 11x slower, per a RapidOCR issue. A NotImplementedError is reported with PaddleOCR 3.7.0 on PaddlePaddle 3.3.1. | Token-loop hallucinations on 5-6 of about 60 Russian/Kazakh scans. Weak on Urdu (43.2%), Sanskrit (51.1%) and old scans (51.1 on olmOCR-bench). One reviewer measured 66 seconds per page on an RTX 4090 with the Hugging Face path. |
Sources
- PaddleOCR GitHub README
- RapidOCR issue #743 denormal weights
- Chandra GitHub README
- Datalab pricing
- GitHub issue #51 token loops
- Medium local test
What each one is
PaddleOCR (PP-OCR)
Baidu's classic open-source text detection and recognition pipeline. Light enough to run on CPU.
PaddleOCR (PP-OCR) detailsChandra
Datalab's OCR model that keeps layout and handles forms, tables and handwriting.
Chandra 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": "paddle/paddleocr","fallbacks": ["datalab/chandra"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
PaddleOCR (PP-OCR) alternativesChandra alternativesBest OCR models in 2026