GoKawiilApple releases LensVLM-9B, a vision-language model that compresses text as images
Apple's AI research team has published LensVLM-9B, a 9-billion-parameter vision language model that processes long documents by first compressing text into image form and then selectively decompressing only the pages it judges relevant, using learned tools. The model, code and demo scripts have been released on GitHub and Hugging Face under Apple's Machine Learning Research Model License and Sample Code License, with configurable compression ratios of 5x, 10x, or 15x.
GoKawiil's interpretation of the reporting above, not reported fact.
Treating long documents as compressed images rather than raw token sequences could let models handle much longer context windows without the memory and compute costs of processing every token directly. Because Apple is releasing this as open research with weights and code, it may let outside researchers test whether selective 'zooming in' on relevant pages is a viable alternative to conventional long-context transformer approaches.
- LensVLM-9B compresses text into images and expands only relevant portions when answering questions
- Apple released model weights and code under separate research and sample-code licenses
- Users can choose compression ratios of 5x, 10x, or 15x when running the model on custom documents
Source: huggingface.co, 2026-09-23
Published there as: “LensVLM: Compressing long context as images, expanding only relevant pages”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.