LensVLM-9B
Apple's first open vision-language model, built on Qwen3.5-9B, for efficient long-document understanding via selective decompression.
Overview
LensVLM-9B is a 9-billion-parameter vision-language model that Apple published on Hugging Face on September 21, 2026, with companion code landing on GitHub the next day -- the company's first open VLM release. Fine-tuned from Alibaba's Qwen3.5-9B, it processes documents as compressed images of text and uses learned tools to selectively expand only the relevant regions back to full resolution instead of decompressing every page, at configurable compression ratios of 5x, 10x, and 15x. The model targets image-text-to-text tasks over long, image-heavy documents, and ships under the Apple Machine Learning Research Model License with sample code under the separate Apple Sample Code License.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- 9B parameters, fine-tuned from Qwen3.5-9B
- Selective decompression: expands only relevant document regions instead of full-page decoding
- Configurable compression levels: 5x, 10x, 15x
- Image-text-to-text architecture for long-document question answering
- Apple's first openly released vision-language model
Use cases
- Question-answering over long, image-heavy documents
- Efficient document analysis without full decompression of every page
- Research and fine-tuning starting point for compressed-context VLM work
Pricing
free, open-weight release under the Apple Machine Learning Research Model License (sample code under the separate Apple Sample Code License)
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 21 Sept 2026
- Record updated
- 28 Sept 2026
- Record last checked
- 28 Sept 2026