LensVLM

LensVLM-9B is a 9B-parameter vision-language model designed to process compressed images of text by selectively expanding relevant regions to their uncompressed form. By utilizing a learned tool-based approach, it maintains high accuracy in document and code understanding tasks while significantly reducing the visual token count required for processing.

LensVLM-9B is a 9B-parameter vision-language model designed to process compressed images of text by selectively expanding relevant regions to their uncompressed form. By utilizing a learned tool-based approach, it maintains high accuracy in document and code understanding tasks while significantly reducing the visual token count required for processing.

• 01Selective context expansion for dynamic high-resolution zooming
• 02Efficient processing of compressed visual text representations
• 03Built on the robust Qwen3.5-9B-Base architecture
• 04Supports multimodal document and code understanding
• 05Optimized for high compression ratios up to 10.1x