docTR vs InternVL3.5
Pick docTR when you need word-level text with coordinates from invoices and receipts, and you want a small Python library you can train yourself. Pick InternVL3.5 when you want a vision language model that outputs Markdown or JSON and can answer questions about a document. The main split is output: docTR returns words, lines and boxes only, while InternVL3.5 gives no boxes or layout by default.
docTRmindee/doctr | 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 | 0.9 s / page | 3.0 s / page |
| Best for | Word-level OCR with positions | Self-hosted document understanding |
| Inputs | PDF, images | images |
| Languages | Mainly Latin scripts | 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 | No | ✓ 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 docTR if
- You need word-level OCR with coordinates, exported as nested page, block, line and word JSON.
- You run on CPU or a single NVIDIA GPU and want ready Docker images, with CUDA 12.2 or newer on the host for the GPU ones.
- You are a Python ML team that wants to choose detection and recognition architectures, with straight or rotated boxes, and train your own models.
Choose InternVL3.5 if
- You need document question answering or Markdown and JSON output from one model.
- You want to trade accuracy for cost across nine sizes, from 1B to 241B-A28B parameters.
- You can use GPUs: 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
| docTR | InternVL3.5 | |
|---|---|---|
| What it is | Two-stage text detection and recognition library from Mindee, built on PyTorch | Open-weight vision language model family from OpenGVLab, 1B to 241B-A28B parameters |
| Licence | Apache-2.0 | Apache-2.0, built on Apache-2.0 Qwen3 |
| Output formats | JSON, plain text, bounding boxes | Plain text, Markdown, JSON |
| Coordinates and layout | Word, line and block coordinates, straight or rotated boxes | No bounding boxes or layout output by default |
| Tables and Markdown | No table, formula or Markdown output; words and lines only | Markdown output; no table-specific claims published |
| Hardware | CPU or NVIDIA GPU; GPU Docker images need CUDA 12.2 or newer | Up to 30B on one A100; 38B needs two A100s; 241B-A28B needs eight A100s |
| Known issues | Cropping images gives poorer results (issue #2007); split-crop merging can misjoin repeated characters (issue #2229); ONNX export of MASTER truncates decoding steps (issue #2224) | OCR scores are published only as chart images, not text values; larger sizes are not suited to CPU-only deployment |
Sources
What each one is
docTR
Open-source OCR library from Mindee with separate text detection and recognition models.
docTR 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": "mindee/doctr","fallbacks": ["opengvlab/internvl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
docTR alternativesInternVL3.5 alternativesBest OCR models in 2026