Hunyuan OCR vs InternVL3.5
Pick InternVL3.5 if you want an Apache-2.0 model you can run at several sizes for document question answering, or if you ship in the EU, UK or South Korea. Pick Hunyuan OCR if you need a roughly 1B model that returns bounding boxes and handles parsing, text spotting, extraction and image translation in one pass. The main split is licence and output: Hunyuan OCR is OCR-specific with coordinates but carries a community licence with regional exclusions, while InternVL3.5 is a general VLM under Apache-2.0 with no bounding boxes by default.
Hunyuan OCRtencent/hunyuan-ocr | InternVL3.5opengvlab/internvl | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 1.5 s / page | 3.0 s / page |
| Best for | Chinese documents and photos | Self-hosted document understanding |
| Inputs | PDF, images | images |
| Languages | 100+ languages | Multilingual |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | No |
| 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 Hunyuan OCR if
- You need bounding boxes, plus Markdown, HTML, LaTeX or JSON output, from one ~1B model.
- You work with photo and scene text, text-image translation, or low-resource and ancient scripts, which are among its 12 task types.
- You want to run on a laptop or CPU through the llama.cpp GGUF path.
Choose InternVL3.5 if
- You are in the EU, UK or South Korea, or need an OSI-approved licence, since InternVL3.5 is Apache-2.0.
- You want to pick a size from nine options, 1B to 241B-A28B, to trade accuracy against cost.
- Your job is document question answering inside a self-hosted multimodal pipeline, and word coordinates are not needed.
Key differences
| Hunyuan OCR | InternVL3.5 | |
|---|---|---|
| Licence | Tencent Hunyuan Community License. Excludes the EU, UK and South Korea. Separate licence needed above 100M monthly active users. | Apache-2.0 |
| Model type and size | OCR-focused end-to-end VLM, about 1B parameters | General VLM family, 1B to 241B-A28B parameters (the 8B variant is 8.5B) |
| Output formats | Markdown, HTML, LaTeX, JSON, plain text, bounding boxes | Plain text, Markdown, JSON. No bounding boxes or layout output by default. |
| Hardware | NVIDIA GPU with CUDA 13 recommended for the unified vLLM setup. llama.cpp GGUF path for CPU, consumer GPU or laptop. | Up to 30B runs on a single A100. 38B needs two A100s. 241B-A28B needs eight A100s. |
| Benchmarks | ParseBench-2 overall 67.11 and ParseBench mean 63.86 (v1.5), with a text score of 62.88 in public CI replay | Not published as text values. The model card shows OCR scores only as chart images. |
| Known failure | The unified environment needs CUDA 13 and vLLM 0.25.1+. An archived 1.0 note said Transformers and vLLM outputs differed in accuracy. | Larger sizes cannot run CPU-only. The largest model needs eight A100s. |
| Price | Free, self-hosted (you pay for compute) | Free, self-hosted (you pay for compute) |
Sources
- HunyuanOCR model card
- Hunyuan Community License
- HunyuanOCR GitHub
- Hugging Face model card (8B)
- InternVL GitHub
What each one is
Hunyuan OCR
Small end-to-end OCR model from Tencent covering parsing, extraction and photo translation.
Hunyuan 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": "tencent/hunyuan-ocr","fallbacks": ["opengvlab/internvl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Hunyuan OCR alternativesInternVL3.5 alternativesBest OCR models in 2026