Baidu Releases Unlimited-OCR, a 3B Parameter Document Parsing Model Based on Deepseek-OCR
Baidu has released Unlimited-OCR, a 3 billion parameter model for optical character recognition and document parsing. The model supports single-page and multi-page document processing with a 32,768 token context window and runs on NVIDIA GPUs using bfloat16 precision.
Baidu Releases Unlimited-OCR, a 3B Parameter Document Parsing Model Based on Deepseek-OCR
Baidu has released Unlimited-OCR, a 3 billion parameter optical character recognition model designed for document parsing. Released on June 22, 2025, the model builds on Deepseek-OCR and is available on Hugging Face.
Technical Specifications
Unlimited-OCR operates with a 32,768 token context window and uses bfloat16 precision on NVIDIA GPUs. The model requires PyTorch 2.10.0, transformers 4.57.1, and CUDA 12.9. According to Baidu, the model is positioned as "pushing Deepseek-OCR one step further" with support for "one-shot long-horizon parsing."
The model offers two processing modes:
- Gundam mode: 1024 base size, 640 image size with crop mode enabled for single images
- Base mode: 1024 base and image size without cropping for single images and all multi-page documents
Deployment Options
Unlimited-OCR can be deployed via Hugging Face transformers or SGLang server infrastructure. The SGLang deployment requires FlashAttention 3 backend and supports an OpenAI-compatible API with streaming responses.
For multi-page documents and PDFs, the model converts pages to images at 300 DPI before processing. The implementation includes custom logit processors with a 35-token n-gram constraint and configurable window sizes (128 tokens for single images, 1,024 tokens for multi-page documents).
Technical Implementation
The model uses PyMuPDF for PDF-to-image conversion and supports both single-image and multi-page inference. Base64-encoded images are sent to the model with text prompts like "document parsing" or "Multi page parsing." The SGLang server configuration allocates 80% of GPU memory statically and disables overlap scheduling.
Baidu acknowledges Deepseek-OCR, Deepseek-OCR-2, and PaddleOCR in the model documentation. Pricing information has not been disclosed.
What This Means
Unlimited-OCR adds another option to the OCR model landscape, though it remains unclear how performance compares to existing solutions like GPT-4V or Claude 3.5 Sonnet on document understanding tasks. The 3B parameter size suggests efficient inference, but no benchmark scores have been published. The model's value proposition depends on comparative accuracy data that Baidu has not yet provided.
Related Articles
Mistral's 3B-Parameter Shieldstral Matches 20B Safety Model on Text Benchmarks
Mistral's new Shieldstral, a 3-billion-parameter open-weight safety classifier, posts an 84.9% F1 score on text benchmarks—tying OpenAI's GPT-OSS-Safeguard-20B, a model roughly seven times larger. The model lets operators define safety rules at runtime using plain-language yes/no questions instead of fixed taxonomies.
Mistral AI Releases Shieldstral-1.0-3B, a 3B-Parameter Policy-Adaptive Safety Classifier
Mistral AI has released Shieldstral-1.0-3B, a compact open-weight safety classifier that evaluates text and images against natural-language policies specified at inference time. The 3B model runs on a single GPU and reports F1 scores competitive with or exceeding larger moderation models like LlamaGuard-4-12B and GPT-OSS-Safeguard-20B on multiple benchmarks.
Black Forest Labs Launches FLUX 3 Video, Claims It Beats Seedance 2.0 on Elo Rankings
Black Forest Labs has made FLUX 3 Video generally available via its API, offering up to 20-second HD/Full HD clips with native audio and lip-sync in 14+ languages. The company claims its internal Elo benchmarks put the model ahead of Seedance 2.0, Gemini Omni Flash, and Minimax H3.
NVIDIA Releases Nemotron VoiceChat 11B, an Open Full-Duplex Speech Model with Live Tool Calling
NVIDIA has released NemotronLabs VoiceChat 11B, an 11-billion-parameter end-to-end full-duplex speech model that unifies streaming speech understanding and generation in one architecture. The model claims to be the first open full-duplex system to support live tool calling during natural conversation, with ~450ms turn-taking latency.
Comments
Loading...