typhoon ocr1.5 2b EU AI Act Compliance Profile

Typhoon-OCR-1.5-2B

A Smaller, More Robust, and Faster Vision-Language OCR for Thai Real-World Documents We’re thrilled to announce Typhoon OCR v1.5, the next evolution of our open-source vision-language document parsing model for English and Thai. Built on top of Qwen3-VL 2B, this release delivers faster inference, improved understanding of handwritten and form-based documents, and enhanced handling of both text-rich and image-rich pages—all in a smaller, more efficient package.

Try our demo available on Demo

Code / Examples available on Github

Release Blog available on OpenTyphoon Blog

Technical Report available on Arxiv

*Remark: This model is intended to be used with a specific prompt only; it will not work with any other prompts.

*Remark: If you want to run the model locally, we recommend using the Ollama build at https://ollama.com/scb10x. We’ve found that the GGUF files for llama.cpp or LM Studio may suffer from accuracy issues.

Key Enhancements:

• Compact and Efficient Architecture: The new version is based on Qwen3-VL 2B, making it significantly smaller while retaining strong multimodal capabilities. Combined with quantization optimizations, Typhoon OCR v1.5 runs efficiently even on lightweight hardware.
• Faster Inference Without PDF Metadata: Unlike Typhoon OCR v1, which relied on embedded PDF metadata for layout reconstruction, v1.5 achieves high layout fidelity directly from image only, eliminating the dependency on metadata. The result: much faster inference across both PDFs and images, without compromising structural accuracy.
• Simplified Single-Prompt Inference: Typhoon OCR v1.5 introduces a single-prompt architecture, replacing the two-prompt process used in v1. This change simplifies integration, reduces complexity in prompt design, and provide

Your obligations as a typhoon ocr1.5 2b deployer

Track your progress

Sign up to track obligations, receive deadline reminders, and generate audit packages.

Tags

Languages

Library

Base Model