Pay per second while your model runs
No subscription required. Buy credits, deploy, and the meter only runs from the moment your model is ready until your idle timeout ends.
Private endpoints
NVIDIA L4, 24 GB
7 to 14B quantized LLMs, any embedding model
- Per hour
- $0.86
- Per second
- $0.00024
- Always on
- $565/mo
NVIDIA L40S, 48 GB
14 to 32B quantized, 8 to 14B full precision
- Per hour
- $2.48
- Per second
- $0.00069
- Always on
- $1,629/mo
Coming soon: quarter-GPU slices for small models at $0.61 per hour.
Plans
Every plan uses the same prices. Pro and Business include credits each month, more endpoints and bonus credit on top-ups.
Pay as you go
$0 / month
$1 trial credit after a card check
- 2 private endpoints
- 50 GB model storage
- Docs and community support
Pro
$29 / month
$29 of credits every month
- 10 private endpoints
- 10% bonus on top-ups
- 200 GB model storage
- Email support
Business
$299 / month
$299 of credits every month
- 50 private endpoints
- 15% bonus on top-ups
- 1 TB model storage
- Priority cold starts
- Shared Slack support
Good to know
- Minimum top-up is $10. Credits from top-ups last 12 months.
- Batch jobs that can wait cost 30% less within an hour and 50% less within a day.
- Priority capacity is 1.5 times the rate and guarantees a GPU when spare capacity is short.
- The first 50 GB of model storage is free, then $0.05 per GB each month.
- If a request fails because of us, it is refunded automatically.
An example month
A chatbot on an L4 that is busy 3 hours a day, including its idle timeout:
$77.76
90 hours at $0.86 per hour. The same GPU running all month would cost $630.72.