pricing.
pay for the model you own. cloud plans include GPU hours and stop at the limit, so a bill never surprises you; add hours when you need them. payments aren't switched on yet: ask for a plan and we'll set it up with you.
cloud
cloud solo
$49 a month
your own private okay model in the cloud, for one person.
- a private endpoint only you can call
- 12 GPU hours a month included
- scales to zero between jobs
- no prompts or answers kept
cloud team
$29 per seat · a month
one private model for your whole team, with keys for each person.
- everything in solo
- 6 GPU hours a month per seat, pooled
- two keys per seat
- usage per day
at least 3 seats
cloud dedicated
$1,690 per GPU · a month
a GPU that's always on and only yours: no cold starts.
- always warm
- unlimited use of your GPUs
- keys for everyone
- priority help
your own hardware
on-prem
$1,500 per GPU · a year
run the okay model on your own servers with NVIDIA GPUs. nothing leaves your network.
- one-line install (docker compose)
- licence checked offline
- a year of updates
- priced per GPU
download
$120 per seat · a year
the okay model on your own Mac. works offline.
- Apple silicon, 32 GB memory or more (16 GB: the mini model)
- licence checked offline
- a year of updates
more hours
10 more GPU hours
$35 once
for a busy month. they don't expire.
what's a GPU hour?
the time your model spends answering: from when a request reaches your endpoint until the answer is done. a typical okaydesigner design takes 20–90 seconds, so 12 hours is a few hundred designs a month. between requests a cloud endpoint sleeps and costs nothing; the first request after a sleep takes up to a minute longer while it wakes.