On-device AI leaderboard

Inference engines and models, ranked on the phones and laptops people own. Every rank links to the experiments and raw data behind it.

Planned Not ranked yet

Which vision-language models can a phone or laptop run, and how quickly?

Vision-language models answering questions about images. Filter by engine, or let each model run on its best compatible engine.

Each row
Vision-language model configurations, on an engine. By default each model runs on its best compatible engine for the objective; filter by engine to fix it.
Views
Overall · Desktop · Mobile
Filters
Engine · Hardware · Workload · Quantization
Measures
Answer quality · Image encoding time · Time to first token · Peak memory

Rank by

  • BalancedAnswer quality first, then latency and memory.
  • SpeedTime to first token after an image.
  • MemoryPeak memory with an image in context.

Its methodology comes first. Nothing is ranked here until its experiments are published with their raw data.