LlamaIndex OCRia-1: A New Vision-Language Model for Parsing Document Tables
Melvin Vivas · X video post · 2026-10-22 · 0:52 · 1,514 views · Open on X
Topics: Retrieval-Augmented Generation (RAG), LLM Fundamentals, Industry Trends & Job Market · Level: intermediate
Summary
Melvin Vivas shares LlamaIndex's launch of OCRia-1, a vision-language model (VLM) for document parsing and OCR. It is built mainly for complex tables like balance sheets, insurance schedules and financial filings. LlamaIndex says it sets new state-of-the-art results for accuracy, cost and latency on its own ParseBench and on third-party benchmarks such as olmOCR-bench. It is available now in LlamaParse.
Key points
- OCRia-1 is a frontier VLM from LlamaIndex built for document parsing. Its main focus is complex tables.
- Claimed scores: 87.4 on ParseBench (a new high), and 83.5 overall on olmOCR-bench, with 92.9 on tables.
- LlamaIndex says it leads three third-party benchmarks as well as its own, and improves on accuracy, cost and latency (price-performance).
- Training approach: trained on the hardest tables LlamaIndex could find, and graded on table structure and meaning, not just text accuracy.
- Target uses: business documents where key data lives in tables, such as balance sheets, insurance schedules and financial filings.
- It is live in LlamaParse, which LlamaIndex says has parsed billions of pages.
- Launch offer: up to $3,000 in credits for enterprises for 7 days.
Resources mentioned
- LlamaIndex (@llama_index) on X · website · x.com · free
LlamaIndex's official X account, which posts updates on its RAG/agent framework and document-parsing products. - OCRia-1 · tool · llamaindex.ai · check price
LlamaIndex's vision-language model for document parsing and OCR, focused on accurately extracting complex tables. - LlamaParse · tool · llamaindex.ai · check price
LlamaIndex's document parsing service that turns PDFs and other documents into structured, LLM-ready text for RAG. - ParseBench · dataset · github.com · free
A document-parsing benchmark used by LlamaIndex to measure parsing accuracy.
Also in: Cohere Parse: Pricing vs Parse Bench Score (Melvin Vivas on X · notes) - olmOCR-bench · dataset · github.com · free
A third-party OCR/document-parsing benchmark from the olmOCR project, with separate scores for tables and other content.
Try this
- Try OCRia-1 in LlamaParse on your hardest table-heavy documents, like the creator plans to.
- When comparing OCR or parsing models, check benchmarks like ParseBench and olmOCR-bench, and look at table scores separately.
- Build a RAG pipeline over financial filings or insurance schedules that uses LlamaParse with OCRia-1 to extract tables, and compare answer quality against a basic text-only parser.