Every provider — OpenAI, Anthropic, Google, Groq, and the rest

"Rate limit exceeded" / 429 Too Many Requests — how kRouter keeps the request alive

HTTP 429 Too Many Requests
{"error":{"message":"Rate limit reached for requests","type":"requests","code":"rate_limit_exceeded"}}

What does "rate limit" mean?

You have sent more requests, or more tokens, than the provider allows in the current window. It is temporary: the window rolls and the account recovers on its own.

Why does it happen?

Bursty agent traffic is the usual cause — parallel tool calls and retries stack up quickly. Some providers enforce a per-minute limit on top of a daily quota, so you can be rate limited while your daily budget is nearly untouched.

How do I fix it?

1. 1

   Add more accounts for the provider. kRouter spreads traffic across them, which is the single most effective fix for per-minute limits.

2. 2

   Set the provider's strategy to round-robin so no one account absorbs a whole burst.

3. 3

   Use a combo so overflow lands on a second provider rather than queueing.

4. 4

   If it repeats on the same account, check whether it is really a daily quota — kRouter distinguishes the two and parks a daily-exhausted account until its real reset instead of retrying every few minutes.

Is it really this error?

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

What will not fix it

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

• Retrying in a tight loop.

  Most providers count rejected requests toward the window, so hammering extends the lockout.

• Making requests smaller.

  If the limit is on requests rather than tokens, size changes nothing.

What kRouter does automatically

kRouter backs the account off with an exponential cooldown that grows on each repeat, capped at 5 minutes. It immediately fails over to your next connected account or provider, so the request still completes.