LlamaIndex launches OpenDocRouter, one API for 10 document parsing models
LlamaIndex's OpenDocRouter puts five frontier and five open-source document-to-markdown models behind one parse endpoint with token-based billing.
LlamaIndex launched OpenDocRouter on October 7, 2026. It is a hosted API that turns PDFs, PNGs and JPEGs into markdown using any of 10 models: five frontier models, including Claude Opus 5.5 and GPT-6 Luna, and five open-source parsers, including MinerU2.5-Pro and PaddleOCR-VL-1.6. If you run an invoice or document-analysis pipeline, switching parsers means changing the model value in the request, not writing a new adapter.
The costs vary widely. LlamaIndex estimates $0.80 per 1,000 pages for GPT-6 Luna, $0.86 for MinerU2.5-Pro and $48.82 for Claude Opus 5.5. Unite.AI notes that these figures come from token usage on LlamaIndex's ParseBench documents, so your pages may cost more or less. On the same models page, Opus 5.5 scores 84.20 overall and Luna 71.34. ParseBench is LlamaIndex's own benchmark, and none of our sources shows an independent reproduction.
Other details from the announcement and coverage:
- Synchronous calls handle up to 50 pages. Documents can be up to 500 pages or 50MB.
- Each page runs as its own call with its own status. AlphaSignal reports that this lets you retry only failed pages, and failed pages are not charged.
- Setting
layout: truereturns bounding boxes with the same layout classes for every model, at an extra $0.20 per million tokens. - According to the docs, as reported by Unite.AI, nothing is retained unless you enable caching, which keeps encrypted results for 24 hours.
Sources disagree on pricing for new accounts: the announcement mentions $5 in credits and $10 top-ups, while Unite.AI and AlphaSignal report a $25 minimum top-up, so check the current terms. No source tests handwriting. Run a sample of your own documents before committing.