Apple publishes LensVLM-9B weights for compressed visual text context

Sources: primary, discussion
Summary: Apple published LensVLM-9B on Hugging Face, with the model card created 2026-09-21 and last modified 2026-09-22. The card names Qwen3.5-9B as the base model, so the weights are an Apple fine-tune of a Qwen base rather than an Apple-trained model, released under the Apple Machine Learning Research Model License. The model treats long text context as compressed images and expands only the pages it needs.
Why it matters: The model reads text as compressed images and expands only the pages it needs, and the Qwen3.5-9B fine-tune carries the Apple Machine Learning Research Model License rather than the base model's own terms.