Supported models
The models in the SophyAI catalog with engine, hardware, measured hourly cost and speed: on your Mac or on cloud GPUs you start.
Each catalog entry is a recipe: model + inference engine + hardware. We measure cost and speed ourselves on Vast.ai (October 2026) and they move with the market: before every start the app shows the price of the actual offer. GPU cost is separate from the app license.
How to read the cards
- $/h: GPU cost on Vast.ai as we measured it, license excluded.
- tok/s: tokens generated per second on a single request.
- —: not measured yet.
Loading the catalog…