Start a fresh conversation, or let your tool compact the history. Attach a narrower slice of the file rather than the whole thing. Route large-context turns to a provider with a bigger accepted payload — kRouter publishes each model's real context_length on /v1/models so your client can pick correctly. Turn on a token saver (RTK or Headroom) to compress the prompt before it is sent.

Retrying, or switching to another account on the same provider. — The request is what is oversized. Every account rejects it identically — which is why kRouter deliberately does not burn your other credentials on it. Raising max_tokens. — That grows the request. It makes this error more likely, not less.

Kiro (AWS Q Developer)

Kiro CONTENT_LENGTH_EXCEEDS_THRESHOLD — "Input content length exceeds threshold"

Matches these upstream messages

content_length_exceeds_threshold
input content length exceeds

What you actually see

{"message":"Input content length exceeds threshold.","reason":"CONTENT_LENGTH_EXCEEDS_THRESHOLD"}

What does "content_length_exceeds_threshold" mean?

The request body itself is larger than the provider will accept. The problem is the request, not the account — which matters, because it means switching accounts cannot help.

Why does it happen?

Usually an agent that has accumulated a long conversation, pasted a large file, or attached a big diff. Kiro's threshold is lower than the model's advertised context window, so you can hit it well before the model would run out of room.

How do I fix it?

Start a fresh conversation, or let your tool compact the history.
Attach a narrower slice of the file rather than the whole thing.
Route large-context turns to a provider with a bigger accepted payload — kRouter publishes each model's real context_length on /v1/models so your client can pick correctly.
Turn on a token saver (RTK or Headroom) to compress the prompt before it is sent.

Is it really this error?

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

Looks like: The model's context window being full.

Actually: The provider's accepted payload is smaller than the model's advertised window, so you can hit this while the model still has room.

Looks like: A rate limit, since it also arrives mid-session as the conversation grows.

Actually: It is deterministic: the same request fails every time, on every account, at any hour.

What will not fix it

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

Retrying, or switching to another account on the same provider.
The request is what is oversized. Every account rejects it identically — which is why kRouter deliberately does not burn your other credentials on it.
Raising max_tokens.
That grows the request. It makes this error more likely, not less.

What kRouter does automatically

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