Baidu’s Unlimited OCR uses R-SWA to parse long documents in one pass
vista8 · x · 2026-07-22
Baidu’s Unlimited OCR is getting strong attention after its release.
- The model uses a Reference Sliding Window Attention (R-SWA) mechanism instead of page-by-page OCR plus result stitching.
- It keeps KV cache size constant during decoding, inspired by how humans read and copy long documents.
- The model has 3B parameters, but inference reportedly activates only 500M parameters and can run locally.
- It can process long documents in one pass, preserving text, formulas, tables, and reading order across pages.
- Reported numbers from the post: 93% accuracy on a benchmark, <0.11 error rate even beyond 40 pages, 2.24M Hugging Face downloads, and 16.7k GitHub stars.
Related event: Baidu Open-Sources Unlimited-OCR for Long Documents(3 posts)→
More from Models
- Engram's random reads don't suit SSDs; CPU-memory over NVLink could serve all 72 GPUs — bookwormengr · 2026-09-11
- Meta's Muse Agent has built-in invite code logic, hinting at free-usage expansion — testingcatalog · 2026-09-11
- Same Echo Maze prompt, three frontier models: all passed visually but shipped the same hidden bug — eyishazyer · 2026-09-11
- Benchmark scores drop from 89% to 19% on new evals — how benchmaxxing breaks leaderboard trust — airesearch12 · 2026-09-11
- ChatGPT tells user their question is too hard and to 'accept dumber answers' — phido3000 · 2026-09-11
- Claude is no longer available for minors as Anthropic rolls out age assurance — Muhammad523 · 2026-09-11