Qwen-VL vs InternVL3.5
Pick Qwen-VL when you need bounding boxes, layout-aware HTML or Markdown output, or OCR across 32 languages, and add loop checks on low-quality scans. Pick InternVL3.5 for self-hosted document question answering when you want a single model family from 1B to 241B-A28B and do not need word coordinates. The main split is output: Qwen-VL returns layout positions, InternVL3.5 returns text, Markdown and JSON only.
Qwen-VLqwen/qwen-vl | 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 | 2.0 s / page | 3.0 s / page |
| Best for | Flexible extraction with prompts | Self-hosted document understanding |
| Inputs | PDF, images | images |
| Languages | 30+ languages | Multilingual |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | 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 Qwen-VL if
- It outputs bounding boxes and a Qwen HTML document format, so you get layout positions and not only text.
- It reads OCR in 32 languages, up from 10 in the previous generation.
- You can also call it as a hosted API on Alibaba Cloud, not only self-host the Apache-2.0 weights.
Choose InternVL3.5 if
- It fits document question answering well, and models up to 30B run on a single A100.
- It comes in nine sizes from 1B to 241B, including MoE variants 20B-A4B and 30B-A3B that activate few parameters per token.
- It has no reported repetition-loop issues in the material reviewed, while Qwen-VL has open reports of looping on low-quality scans.
Key differences
| Qwen-VL | InternVL3.5 | |
|---|---|---|
| Licence | Apache-2.0 | Apache-2.0, built on Apache-2.0 Qwen3 |
| How you run it | Self-host the open weights, or use the hosted API on Alibaba Cloud | Self-host only, with Transformers, LMDeploy or vLLM |
| Price | Free self-hosted (you pay for compute). Hosted pricing varies by model and was not verified. | Free self-hosted (you pay for compute) |
| Output formats | Plain text, HTML, bounding boxes, JSON, Markdown | Plain text, Markdown, JSON. No bounding boxes or layout output by default. |
| Model sizes | 2B, 4B, 8B, 32B dense; 30B-A3B and 235B-A22B MoE | 1B to 241B-A28B; the 8B variant is 8.5B |
| Benchmark | Qwen 3 VL 8B scores 64.6 on olmOCR-bench (independent Datalab table), below dedicated OCR models | Not published as text values; the model card shows OCR scores only as chart images |
| Known failure | Can loop and repeat the same text, worse with greedy decoding and on low-quality scans. Qwen2.5-VL-32B GGUF often loops on scanned documents. | Not published. Largest 241B-A28B needs eight A100 GPUs, and 38B needs two A100s. |
Sources
- Qwen3-VL GitHub
- Issue #1611 infinite repetition
- llama.cpp issue #14663
- Chandra README benchmark table
- Hugging Face model card (8B)
- InternVL GitHub
What each one is
Qwen-VL
Alibaba's open-weight vision language models (Qwen2.5-VL, Qwen3-VL). Strong general OCR and grounding.
Qwen-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": "qwen/qwen-vl","fallbacks": ["opengvlab/internvl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qwen-VL alternativesInternVL3.5 alternativesBest OCR models in 2026