Live
One model name: auto
Run switchboard serve, point your app's OpenAI client at it, and send model="auto". Every request goes to Surplus with your key, routed to the cheapest model that can handle it.
The Surplus launchpad, live
Builders launch open models on Surplus, each on its own GPU, and earn every time someone calls them. Here is every model live on the launchpad, how to call it, and how Switchboard sends it traffic.
Live on the Surplus API
| Model | Size | Built on | Price | Launched | Status | Switchboard |
|---|---|---|---|---|---|---|
| Loading the launchpad… | ||||||
Send them traffic
Live
Pick a launch model per task
Send one kind of prompt to a launchpad model, for example easy code questions to a coder. If its GPU is asleep, Switchboard waits for it, then falls back so your app never sees the error.
Next
A report card for every launch
Switchboard will test each launch model on prompts that match what it was built for, and publish the method with the results. Models that pass get traffic from Switchboard by default. New launches are graded as they arrive.
$ pip install switchboard-router
$ switchboard --launchpad # launch models you can call right now
$ switchboard serve --override small/code=launch/0XARTEX/artex-coder-7b@9182fd6Until the report card is out, Switchboard sends traffic to a launch model only when you ask for it with --override. Try it on your own prompts first.