Retry shortly — capacity errors usually clear in seconds to minutes. Route to a different model or provider for the moment; a combo does this automatically instead of failing the request. Prefer a less contended model for background work so your agent keeps moving when the flagship is saturated.

Adding more accounts on the same provider. — The shortage is theirs, not per-account. Every account of theirs sees the same wall. Retrying instantly in a loop. — It adds load to an already saturated pool. A short pause, or a different model, resolves it faster.

Anthropic, Google, and other capacity-constrained providers

"No capacity available" / "Overloaded" — provider is out of compute

Matches these upstream messages

no capacity available
capacity
overloaded

What you actually see

{"error":{"type":"overloaded_error","message":"Overloaded"}}

What does "no capacity available" mean?

The provider has no free capacity for that model right now. It is their infrastructure, not your account — an identical request from anyone else would fail the same way.

Why does it happen?

Peak demand on a popular model, or a regional shortage. Newly launched models are the most affected.

How do I fix it?

Retry shortly — capacity errors usually clear in seconds to minutes.
Route to a different model or provider for the moment; a combo does this automatically instead of failing the request.
Prefer a less contended model for background work so your agent keeps moving when the flagship is saturated.

Is it really this error?

These get mistaken for each other, and chasing the wrong one is where the time goes.

Looks like: A rate limit on your account.

Actually: Capacity errors are about the provider's hardware, not your usage. They hit accounts that have sent nothing all day.

Looks like: A network failure.

Actually: You got a valid HTTP response — the provider answered, it just had nowhere to run the request.

What will not fix it

Reasonable-sounding fixes that do not apply here — worth ruling out before you spend an afternoon on them.

Adding more accounts on the same provider.
The shortage is theirs, not per-account. Every account of theirs sees the same wall.
Retrying instantly in a loop.
It adds load to an already saturated pool. A short pause, or a different model, resolves it faster.

What kRouter does automatically

kRouter parks that model on that account for 5 seconds. It does NOT try your other accounts, because the same request would fail identically on all of them — the error is returned to you straight away instead of burning every credential you have.

Stop hitting this error

kRouter routes your AI coding tools across every provider and account you own, so a limit on one does not stop your work. It is free, MIT-licensed and runs on your machine.

Install kRouter