Unlimited-OCR
Baidu's open vision-language model for one-shot long-horizon document parsing.
About
Unlimited-OCR is Baidu's open-weights vision-language model for document understanding, released in June 2026 under the MIT License with a paper on arXiv (2606.23050) and code on GitHub (baidu/Unlimited-OCR). Its pitch is "one-shot long-horizon parsing": instead of chunking documents page by page through a pipeline, the model takes a document-centric view and parses long documents in a single pass, aimed at RAG ingestion, PDF conversion, and table extraction workloads. It runs in Transformers, gained vLLM inference support from the community shortly after release, and is also served on Baidu Cloud, with an interactive demo on Hugging Face Spaces.
As with most document-VLM releases, the headline capability framing is Baidu's own and has not been independently evaluated for this entry; what is verified is the checkpoint, the MIT license, the paper, the vLLM support, and the Baidu Cloud and Spaces availability. The repository shows active community uptake (ms-swift training support added within weeks of release), which is a reasonable signal for a model this new.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via Hugging Face model card (baidu/Unlimited-OCR): MIT license, 'one-shot long-horizon parsing' document VLM, vLLM + ms-swift support, Baidu Cloud availability, HF Spaces demo, paper arXiv 2606.23050, code at github.com/baidu/Unlimited-OCR.; Weights confirmed downloadable on HF (ungated); repo created 2026-06-19; 4.7k likes; release timeline documented in card (paper 2026-06-23, demo 2026-06-24, vLLM 2026-06-28).; Capability framing is Baidu's own; no independent evals consulted. Active community uptake (ms-swift, vLLM, Baidu Cloud within weeks) is a positive freshness signal.. License: MIT (model card frontmatter).