GPT vs Chandra
Pick GPT when you need handwritten notes, extraction into custom JSON, or document question answering through a hosted API. Pick Chandra when you need PDF to Markdown with bounding boxes, multilingual scans, or the option to self-host. The main split is that Chandra gives layout output and a self-hosting option, while GPT gives flexible prompting but no coordinates.
GPTopenai/gpt | Chandradatalab/chandra | |
|---|---|---|
| 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 | 2.0 s / page |
| Best for | Messy scans, custom instructions | Complex layouts and handwriting |
| Inputs | PDF, images | PDF, images |
| Languages | 50+ languages | 40+ languages |
| Max pages per request | 100 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | ✓ Yes |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | No |
| Forms | No | ✓ Yes |
| 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 GPT if
- It is reported to handle messy cursive well: an HN user saw about 5% errors versus about 30% for Mistral OCR.
- You can prompt it to extract data into custom JSON and to answer questions about a document, not only transcribe it.
- The API takes files up to 50 MB per request and sends both extracted text and page images to the model for PDFs.
Choose Chandra if
- It returns Markdown, HTML or JSON plus layout chunks with bounding boxes and labels per block, which GPT's standard response does not.
- You can self-host the open weights on a single NVIDIA GPU, or use the hosted Datalab API at $1, $3 or $7.50 per 1,000 pages.
- It scores 85.8 on olmOCR-bench and averages 77.8% across 43 languages, and it is aimed at tables, forms and math.
Key differences
| GPT | Chandra | |
|---|---|---|
| Deployment | Hosted OpenAI API only | Self-host (vLLM or Hugging Face) or hosted Datalab API |
| Licence | Proprietary | Code Apache-2.0; weights modified OpenRAIL-M, free under $2M funding/revenue, commercial licence needed above that |
| Price | Token-based per model; page images add input tokens, so cost per page is not fixed | $1, $3 or $7.50 per 1,000 pages on the hosted API; self-hosted is free within licence terms plus your compute |
| Output formats | Markdown, JSON, plain text; no bounding boxes or confidence scores in the standard response | Markdown, HTML, JSON and bounding boxes |
| Benchmark | OmniDocBench v1.5 overall 85.50 for GPT-5.2 (independent); 86.59 on v1.6 | olmOCR-bench overall 85.8 for Chandra 2 (reported by the authors); different benchmark, not directly comparable |
| Handwriting | One HN report of about 5% errors on messy cursive | Not published |
| Known failure | Cost per page varies with page images; no word coordinates | Token-loop hallucinations reported on 5-6 of about 60 Russian/Kazakh scans; weak on Urdu (43.2%), Sanskrit (51.1%) and old scans (51.1 on olmOCR-bench) |
Sources
- OpenAI file inputs docs
- OmniDocBench leaderboard
- HN Mistral OCR thread
- Chandra GitHub README
- Datalab pricing
- GitHub issue #51 token loops
What each one is
Chandra
Datalab's OCR model that keeps layout and handles forms, tables and handwriting.
Chandra detailsTry 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": ["datalab/chandra"],"file_url": "https://example.com/document.pdf"}