Skip to content
Docs

Free Tier

A free tier for non-commercial and evaluation use.

Free tier
Concurrent inference servers 2
Nodes 1
Confidential computing not included
Everything else included

“Everything else” is not a euphemism: the full optimization layer, all four routing modes, the load balancer and its algorithms, the capacity gate and queue, the complete API surface, the model registry, roles, the audit log and the console are all present.

Three different free windows exist and are easy to conflate:

Grace period Free tier Paid trial
Length 24 hours indefinite 7 days
Starts on first boot activating a free licence first paid subscription
Needs nothing a free licence key a card
Limits none 2 servers, 1 node none

The grace period is the unlicensed window every install gets on first boot, with no limits — install and serve immediately. The free tier is a licence with limits that does not expire.

Starting a third server is refused with a clear message naming the limit, not a generic failure. Servers already running keep running.

The server being started is excluded from the count, so restarting an already-running server does not fail against its own limit.

Worth knowing if you are on a newer plan than an older build understands: an entitlement a build does not recognise is treated as unlimited, not as free. Somebody paid for something newer than the image knows about, and degrading them out of the build’s own ignorance is the worse failure.