OCR for long documents in a single pass
A post published on June 23, 2026 describes an OCR model that processes documents longer than 40 pages in one pass while keeping KV Cache size constant.
Source: Unlimited OCR para documentos longos (github.com). Text prepared with AI from this source.
What happened and what to do
On June 23, 2026, Baidu published a post about an open-source OCR model for long documents. Its approach uses Reference Sliding Window Attention to process more than 40 pages in a single pass, with 3 billion total parameters and 500 million activated. The post also reports results on OmniDocBench v1.5 and v1.6 and describes constant KV Cache size.
A company can evaluate this approach on its own document collection, starting with a diagnosis of document types and extraction requirements. A pilot can compare OCR quality, memory use, and processing time on long documents; an implementation can then connect extraction to an API or pipeline, with monitoring and failure review. This calls for document-engineering work and evaluation, without assuming a particular model is suitable before testing.
How the consultancy can help
Wendelmaques can diagnose the document workflow and OCR requirements, define an evaluation using the client’s criteria and samples, and propose a scoped implementation covering integration, monitoring, and operation.
Next step
Send a short description of the documents and workflow you want to improve to receive a scoped proposal for diagnosis and implementation.
Let's assess your project
I reply with scope and a proposal.