Baidu open-sources Unlimited OCR 2: a 3B model that parses hundreds of pages in one shot
abhishek__AI · x · 2026-10-27
Baidu released Unlimited OCR 2, a fully open-source (MIT) 3B-parameter OCR model that can parse hundreds of pages in a single request, billed as the start of the "OCR-3.0 / unlimited-horizon parsing" era.
- Uses SR-SWA to keep text-visual attention in check, yielding near-constant decoding latency and virtually unlimited document length
- Benchmark charts show 77.1 on the hundreds-of-pages MarathonDocBenchmark, far ahead of Qwen3.8-Max (22.9) and Gemini-3.1-Pro Preview (27.0); it also tops MPDocBench at 84.6, +6.3 over the runner-up
- Already supports vLLM inference and ms-swift training, with demos on Hugging Face and Baidu Cloud
Code and paper are on GitHub (baidu/Unlimited-OCR).
Related event: Baidu Open-Sources Unlimited OCR 2 for Thousand-Page Document Parsing(21 posts)→
More from Models
- inclusionAI's Ling-3.1-flash-1019: 560B MoE, 25B active, 1M context, AA score up to 44 — jacek2023 · 2026-10-27
- User shows Gemini ignoring explicit custom instructions and starting actions unbidden — loriofficialita · 2026-10-27
- EdgeRock architecture keeps KV cache at 72 MiB for 256K context, 99.1% smaller than Qwen3.5-4B — alifcoder · 2026-10-27
- Dev Who Called Kimi Subscription a Scam Gets Locked Out Over False Paid-Post Flag — Scobleizer · 2026-10-27
- Blogger reverses stance: Anthropic now 'crushes' OpenAI models, opposite of 3 months ago — Scobleizer · 2026-10-27
- Power user's stress test: ChatGPT great at research, flaky at image consistency — andytechuk · 2026-10-27