Skip to content
FastVLM.net

MODEL CATALOG

Choose the right FastVLM model

The three official sizes share the FastViTHD vision encoder strategy but target different latency, footprint and quality priorities.

Fastest

FastVLM-0.5B

The smallest official FastVLM variant and the model behind the browser playground. Start here when latency, footprint and portability matter most.

Best for
Browser, WebGPU and constrained edge devices
Runtimes
Browser · Transformers.js · Python · Apple Silicon
Model details
Balanced

FastVLM-1.5B

A middle-sized official variant that improves every score in Apple’s published table over 0.5B while remaining substantially smaller than 7B.

Best for
Balanced local inference and Apple Silicon development
Runtimes
Python · Apple Silicon · Workstation
Model details
Best quality

FastVLM-7B

The largest official FastVLM variant and the top scorer in every row of Apple’s published evaluation table. Prefer native desktop-class deployment.

Best for
Highest published FastVLM benchmark quality
Runtimes
Python · Desktop · Native deployment
Model details

A practical selection rule

  • Start with 0.5B when you need WebGPU or the lowest footprint.
  • Choose 1.5B when you can spend more memory for consistent benchmark gains.
  • Use 7B when published accuracy matters more than latency and native deployment is available.

Model names describe the LLM class, not the total parameter count shown by every distribution. Check the exact model card and file sizes before provisioning hardware.