PaddleOCR-VL vs InternVL3.5
Pick PaddleOCR-VL for document parsing on a modest GPU: it is under 1B parameters and outputs Markdown and JSON for pages with text, tables, formulas and charts. Pick InternVL3.5 when you need document question answering or one model family at several sizes. The main split is a purpose-built parser against a general vision language model.
PaddleOCR-VLpaddle/paddleocr-vl | InternVL3.5opengvlab/internvl | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 6 credits / page | 2 credits / page |
| Per 1,000 pages | $6.00 | $2.00 |
| Typical speed | 1.3 s / page | 3.0 s / page |
| Best for | CJK documents | Self-hosted document understanding |
| Inputs | PDF, images | images |
| Languages | 109 languages | Multilingual |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ 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-VL if
- It recognizes text, tables, formulas and charts and outputs Markdown and JSON for full pages.
- It supports 109 languages in one model, including Cyrillic, Arabic, Devanagari and Thai scripts.
- At under 1B parameters it runs on much smaller GPUs than 7B document VLMs.
Choose InternVL3.5 if
- It comes in nine sizes from 1B to 241B-A28B, so you can trade accuracy for cost.
- It suits document question answering and self-hosted multimodal pipelines, not only text extraction.
- Models up to 30B fit on a single A100, and the MoE variants 20B-A4B and 30B-A3B activate few parameters per token.
Key differences
| PaddleOCR-VL | InternVL3.5 | |
|---|---|---|
| Licence | Apache-2.0 | Apache-2.0 |
| Model size | 0.9B | 1B to 241B-A28B; the 8B variant is 8.5B |
| Hardware | NVIDIA GPU recommended; CPU possible but slow; no minimum VRAM stated | Up to 30B on a single A100; 38B needs two A100s; 241B-A28B needs eight A100s |
| Output formats | Markdown and JSON | Plain text, Markdown and JSON |
| Layout and coordinates | Layout analysis for full pages; the transformers path does element-level recognition only | No bounding boxes or layout output by default |
| Published OCR benchmarks | OmniDocBench v1.6 overall 96.33 (version 1.6, authors); olmOCR-Bench overall 80.0 (independent) | Not published as text; the model card shows OCR scores only as chart images |
| Known failure | Version 1.6 regresses Russian and Kazakh OCR on born-digital PDFs, hangs on blurry images, and decodes about 11x slower because use_cache ships as false | Not published |
Sources
- PaddleOCR-VL-1.6 model card
- olmOCR-Bench table
- Issue #18088 Cyrillic regression
- Issue #18453 use_cache slowdown
- Hugging Face model card (8B)
- InternVL GitHub
What each one is
PaddleOCR-VL
Open-source document model from Baidu. Excellent on Chinese, Japanese and Korean.
PaddleOCR-VL 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-vl","fallbacks": ["opengvlab/internvl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
PaddleOCR-VL alternativesInternVL3.5 alternativesBest OCR models in 2026