APPLE MODELS · DEMOS · GUIDES
Apple’s research teams publish far more than FastVLM. This hub organises the models by task, marks which ones have an in-browser tool or a tested local guide on this site, and links every entry to its official source.
Vision-language models, image–text encoders and depth estimation that run on-device.
Vision-language model with the FastViTHD encoder; 0.5B runs in your browser.
Metric depth and focal length from one image; powers the depth map generator.
Low-latency image–text encoders; zero-shot classification and similarity tools.
Open vision encoders (0.3B–2.7B) with PyTorch, JAX and MLX backends.
Single-image 3D: nearby-view synthesis and relightable asset generation.
Normalizing-flow generators for images, video and unified multimodal generation.
Research models for code generation and retrieval-augmented reasoning.
7B masked-diffusion language model for code generation.
Retrieval and generation joined through continuous latent reasoning.
Embarrassingly simple self-distillation that improves code generation (Apple Research).
The Apple Foundation Models behind Apple Intelligence and the developer framework that exposes them.
AFM 3 Core, Core Advanced and the Private Cloud Compute models behind Apple Intelligence.
Swift and Python APIs for on-device LLM sessions, guided generation, tools and multimodal prompts.
Depth maps, classification, similarity and image Q&A—each tool names the Apple model behind it.
Browse toolsCompare FastVLM with SmolVLM2, LFM2.5-VL, Gemma and other small VLMs that run on-device.
On-device VLM hub