PaddleOCR (PP-OCR) vs InternVL3.5
Pick PaddleOCR (PP-OCR) for fast text line detection and recognition on printed Chinese and English, receipts, signs and screenshots, especially on CPU, edge or in the browser. Pick InternVL3.5 for document question answering and Markdown or JSON output, when you have GPUs and want one model family at several sizes. The main split is a small detection and recognition pipeline that returns boxes versus a general vision language model that returns text without coordinates.
PaddleOCR (PP-OCR)paddle/paddleocr | InternVL3.5opengvlab/internvl | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 0.7 s / page | 3.0 s / page |
| Best for | Fast printed text, Chinese and English | Self-hosted document understanding |
| Inputs | PDF, images | images |
| Languages | 80+ languages | Multilingual |
| Max pages per request | 500 | 500 |
| Tables | No | ✓ Yes |
| Handwriting | No | No |
| Math | No | ✓ Yes |
| Bounding boxes | ✓ Yes | No |
| Confidence | ✓ Yes | 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 PaddleOCR (PP-OCR) if
- You need text boxes with the recognised text, since it returns plain text, bounding boxes and JSON.
- You run on CPU, edge or mobile, or in the browser: PP-OCRv6 medium is 34.5M parameters, reports a 5.2x CPU speedup with OpenVINO and has an official PaddleOCR.js SDK.
- You read printed Chinese, English or Japanese, or any of 46 Latin-script languages, with one PP-OCRv6 model covering 50 languages.
Choose InternVL3.5 if
- You want to ask questions about a document, not only extract its text, and get Markdown or JSON back.
- You want to pick a size by cost: nine sizes run from 1B to 241B-A28B, and models up to 30B fit on a single A100.
- You want a licence and base model that are both Apache-2.0 (built on Apache-2.0 Qwen3), with MoE variants 20B-A4B and 30B-A3B that activate few parameters per token.
Key differences
| PaddleOCR (PP-OCR) | InternVL3.5 | |
|---|---|---|
| What it is | Lightweight text detection and recognition pipeline from Baidu PaddlePaddle | Open-weight vision language model family from OpenGVLab, 1B to 241B-A28B parameters |
| Licence and price | Apache-2.0, free, self-hosted (you pay for compute) | Apache-2.0, free, self-hosted (you pay for compute) |
| How you run it | paddleocr Python package, C++ or the PaddleOCR.js browser SDK | Transformers, LMDeploy or vLLM |
| Hardware | CPU, NVIDIA GPU, Kunlunxin XPU and other accelerators; 0.13s on A100 for PP-OCRv6 | Up to 30B on one A100, 38B on two A100s, 241B-A28B on eight A100s; larger sizes are a poor fit for CPU-only |
| Output | Plain text, bounding boxes, JSON | Plain text, Markdown, JSON; no bounding boxes or layout output by default |
| Tables and formulas | Not handled by the base pipeline; needs the separate PP-Structure or VL pipelines | Not published |
| Known issues | NotImplementedError on PaddleOCR 3.7.0 with PaddlePaddle 3.3.1 (issue #18162); PP-OCRv6 weights contain denormals that make CPU inference 4 to 11x slower (RapidOCR issue #743) | OCR scores are published only as chart images, not text values |
Sources
- PaddleOCR GitHub README
- PP-OCRv6 on Hugging Face
- RapidOCR issue #743 denormal weights
- Hugging Face model card (8B)
- InternVL GitHub
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) detailsInternVL3.5
Shanghai AI Lab's open vision-language family (1B to 241B, August 2025), strong on OCR and document benchmarks.
InternVL3.5 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": ["opengvlab/internvl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
PaddleOCR (PP-OCR) alternativesInternVL3.5 alternativesBest OCR models in 2026