LensVLM is a 9-billion-parameter vision-language model from Apple's AI/ML Research team, built on Qwen3.5-9B-Base and released with the paper "LensVLM: Selective Context Expansion for Compressed Visual Representation of Text" (arXiv:2605.07019). The project targets a trade-off in VLMs that read rendered text as images: shrinking a document into fewer, denser page images cuts the visual tokens a model must process, but once character strokes fall below the vision encoder's effective resolution, accuracy collapses. LensVLM's fix is an inference framework with a post-training recipe that teaches the model to first scan compressed page images, then call a learned read_page tool to selectively re-render only the pages it judges relevant back into uncompressed, legible form before answering. In the released demo, the model is shown fifteen compressed page images from a HotpotQA-style multi-hop question, reasons inside think tags about which page likely holds the answer, calls read_page on that page, and answers using the retrieved text. Per the paper's abstract, LensVLM keeps accuracy comparable to a full-text upper bound at 4.3x effective compression, and outperforms retrieval-based, text-compression, and visual-compression baselines up to 10.1x effective compression across seven text QA benchmarks; it also generalizes to multimodal document and code-understanding tasks, with its accuracy advantage widening as compression increases. The repository includes the page-rendering pipeline with three compression presets (5x, 10x, 15x), dataset providers that rebuild the paper's HotpotQA, Natural Questions, and Musique evaluation sets by adding distractor paragraphs to gold passages, an evaluation pipeline that scores answers with an LLM-as-judge (the paper serves Qwen3.5-397B-A17B-FP8 as judge via vLLM), and a demo script for running the model on a custom text file. Model weights ship under the Apple Machine Learning Research Model License, while the code is released under the Apple Sample Code License, permitting reuse and modification provided Apple's copyright notice is kept and its name is not used to promote derivative works.