Qwen-VL vs Pixtral
Pick Qwen-VL for mixed OCR plus reasoning, photos and screenshots with text, and new projects, because it is actively developed and reads 32 languages. Pixtral 12B suits question answering over charts and documents, or teams already on the Mistral API. Mistral marks it deprecated, so avoid it for new OCR work.
Qwen-VLqwen/qwen-vl | Pixtralmistral/pixtral | |
|---|---|---|
| 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 | Questions about document images |
| Inputs | PDF, images | images |
| Languages | 30+ languages | Multilingual |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | No |
| 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 does OCR in 32 languages, while Pixtral is reported to lose accuracy on non-Latin scripts.
- It comes in sizes from 2B to 235B-A22B MoE, all Apache-2.0, so you can match the model to your hardware.
- It outputs layout positions, bounding boxes and a Qwen HTML document format, not just plain text.
Choose Pixtral if
- It scores 90.7 ANLS on DocVQA per Mistral, which suits question answering over documents.
- It has a 128K token context, so one request can carry many images.
- It was priced at $0.15 per million input and output tokens on the Mistral API, and existing Mistral integrations keep working.
Key differences
| Qwen-VL | Pixtral | |
|---|---|---|
| Licence | Apache-2.0 | Apache-2.0 |
| Model sizes | 2B, 4B, 8B, 32B dense; 30B-A3B and 235B-A22B MoE | One size: 12.4B (12B decoder + 400M vision encoder) |
| How you run it | Self-host open weights, or Alibaba Cloud hosted API. FP8 checkpoints need an H100-class GPU and CUDA 12+. | Self-host open weights, or the Mistral API as pixtral-12b-2409 |
| Price | Free to self-host. Hosted pricing on Alibaba Cloud: Not published. | $0.15 per million input and output tokens at launch on the Mistral API. Free to self-host. |
| Output formats | Plain text, HTML, bounding boxes, JSON, Markdown | Plain text, Markdown, JSON |
| Benchmark | 64.6 on olmOCR-bench (Qwen 3 VL 8B), run by an independent party, below dedicated OCR models | 90.7 ANLS on DocVQA, 81.8 on ChartQA, 52.5 on MMMU, all reported by Mistral |
| Known failure | Can loop and repeat the same text, mostly on low-quality scans and worse with greedy decoding. Qwen2.5-VL-32B GGUF often loops on scanned documents. | Weaker on handwriting, small fonts and non-Latin scripts. Q4 quantization degrades small text OCR. Deprecated and no longer maintained. |
Sources
- Qwen3-VL GitHub
- Issue #1611 infinite repetition
- Chandra README benchmark table
- Mistral announcement
- Mistral Sept 2024 pricing
- Third-party deployment guide
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 detailsPixtral
Mistral's open-weight vision models, Pixtral 12B and Pixtral Large. Mistral's API no longer serves them, so they run self-hosted.
Pixtral 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": ["mistral/pixtral"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qwen-VL alternativesPixtral alternativesBest OCR models in 2026