OCRRouter

GLM-OCR vs Qwen-VL

Pick GLM-OCR for high-volume PDF to Markdown, academic papers with formulas, and schema-based field extraction, because it is a 0.9B model with a $0.03 per million token API and a 94.62 OmniDocBench v1.5 score. Pick Qwen-VL when OCR is one part of a wider job, such as photos, screenshots, VQA or reasoning over the text, and you want one general model in sizes from 2B to 235B. Watch both for looping output, and check GLM-OCR's table output closely if every row and decimal must survive.

GLM-OCRzai/glm-ocrQwen-VLqwen/qwen-vl
TypeOpen sourceOpen source
Price6 credits / page2 credits / page
Per 1,000 pages$6.00$2.00
Typical speed1.3 s / page2.0 s / page
Best forLow-cost general documentsFlexible extraction with prompts
InputsPDF, imagesPDF, images
Languages100+ languages30+ languages
Max pages per request500200
Tables✓ Yes✓ Yes
HandwritingNo✓ Yes
Math✓ Yes✓ Yes
Bounding boxes✓ Yes✓ Yes
Confidence✓ YesNo
FormsNoNo
Multilingual✓ Yes✓ Yes
On OCRRouterAvailableAvailable

Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.

Which to choose

Choose GLM-OCR if

  • It costs $0.03 per million tokens for input and output on the Z.ai API, and self-hosting is free apart from compute.
  • It is a 0.9B model that reached 1.86 PDF pages per second on a single replica in the authors' test.
  • It can extract fields into a user-defined JSON schema, which suits forms, and it scores 94.62 on OmniDocBench v1.5 per its own model card.

Choose Qwen-VL if

  • It is a general vision-language model, so one deployment can do OCR plus reasoning or VQA on photos and screenshots.
  • It reads text in 32 languages and handles low light, blur and tilted text per Qwen.
  • It outputs bounding boxes, layout positions and a Qwen HTML document format, and comes in 2B to 235B-A22B sizes, all Apache-2.0.

Key differences

GLM-OCRQwen-VL
Model size0.9B2B, 4B, 8B, 32B dense; 30B-A3B and 235B-A22B MoE
LicenceMIT for the weights; the SDK repo and bundled PP-DocLayoutV3 are Apache-2.0Apache-2.0 (Qwen3-VL repository)
PriceZ.ai API: $0.03 per million tokens for input and output. Self-hosted: free plus compute.Self-hosted: free plus compute. Alibaba Cloud hosted pricing varies by model; Not published here.
Output formatsMarkdown, JSON, plain textPlain text, HTML, bounding boxes, JSON, Markdown
BenchmarksOmniDocBench v1.5 94.62 and olmOCR-Bench 75.2 (authors); ParseBench mean 29.6 (independent)olmOCR-bench 64.6 for Qwen 3 VL 8B (independent, Datalab table)
API and hardware limitsAPI caps input at 100 pages, 50 MB per PDF and 10 MB per image. No aarch64 Linux support (issue #227).FP8 checkpoints need an H100-class GPU and CUDA 12+. The 235B example uses 8-way tensor parallel on vLLM.
Known failuresRepeats output unless a stop sequence is set (issue #220). Drops table footer and summary rows (#240). Loses comma decimal separators in cropped tables (#241).Can loop on repeated text, worse with greedy decoding (issue #1611), mostly on low-quality images. Qwen2.5-VL-32B GGUF often loops on scanned documents (llama.cpp #14663).

Sources

What each one is

GLM-OCR

Small open-source OCR model from Z.ai with layout regions. MIT licensed.

GLM-OCR details

Qwen-VL

Alibaba's open-weight vision language models (Qwen2.5-VL, Qwen3-VL). Strong general OCR and grounding.

Qwen-VL details

Try both on your own file

Run the file once with each model and compare the output.

PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?

Use one, fall back to the other

{"model": "zai/glm-ocr","fallbacks": ["qwen/qwen-vl"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

GLM-OCR alternativesQwen-VL alternativesBest OCR models in 2026