GPT vs Docling
Pick GPT when you need handwriting, custom JSON extraction with reasoning, or document question answering through a hosted API. Pick Docling when you ingest born-digital PDFs and Office files for RAG and want a free, MIT-licensed library you run yourself. The main split is a paid hosted model versus a self-hosted converter whose scanned-page accuracy depends on the OCR engine you plug in.
GPTopenai/gpt | Doclingibm/docling | |
|---|---|---|
| Type | Vision model | Open source |
| Price | 11 credits / page | 2 credits / page |
| Per 1,000 pages | $11.00 | $2.00 |
| Typical speed | 2.4 s / page | 2.2 s / page |
| Best for | Messy scans, custom instructions | Structured export for RAG |
| Inputs | PDF, images | PDF, images, DOCX |
| Languages | 50+ languages | Latin scripts, CJK |
| Max pages per request | 100 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | No |
| 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 GPT if
- It handles handwritten notes: one HN user reported about 5% errors on messy cursive versus about 30% for Mistral OCR.
- It can extract data into custom JSON and answer questions about a document, because you prompt it to do so.
- It needs no infrastructure: send PDFs or images up to 50 MB per request to the OpenAI API.
Choose Docling if
- It is free under the MIT license and runs locally on macOS, Linux and Windows, on CPU or GPU, so you pay only for compute.
- It parses PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails and XBRL through one API.
- Its lossless JSON export keeps layout, reading order and table structure, and it has LangChain and LlamaIndex integrations.
Key differences
| GPT | Docling | |
|---|---|---|
| How you run it | Hosted OpenAI API | Python library, runs locally |
| License | Proprietary | MIT (individual models have their own licenses) |
| Price | Token-based per model; page images add input tokens, so cost per page is not fixed | Free, self-hosted; you pay for compute |
| Input formats | Images and PDFs (up to 50 MB per request) | PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails, XBRL |
| Output formats | Markdown, JSON, plain text | Markdown, HTML, JSON, DocTags, plain text |
| Benchmark | GPT-5.2 scored 85.50 on OmniDocBench v1.5 and 86.59 on v1.6 (independent) | Not published |
| Known failure | No bounding boxes or confidence scores in the standard response | OCR regression in 2.143.1 drops words and numbers with full-page Tesseract OCR; reading-order rule merges neighbouring columns on some layouts |
Sources
- OpenAI file inputs docs
- OmniDocBench leaderboard
- HN Mistral OCR thread
- Docling GitHub README
- Supported formats
- Issue #5254 OCR regression
What each one is
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": "openai/gpt","fallbacks": ["ibm/docling"],"file_url": "https://example.com/document.pdf"}