Error messages

This request requires more credits, or fewer max_tokens.

This request requires more credits, or fewer max_tokens.

The first sentence is fixed. The numbers after it are yours:

This request requires more credits, or fewer max_tokens. You requested up to 16000 tokens, but can only afford 15154. To increase, visit https://openrouter.ai/settings/credits and add more credits

It's an HTTP 402. The newer error bodies carry a hint and a source field too, as in this September 2026 report:

"metadata": {
  "limit_source": "openrouter_credits",
  "remedy_hint": "Add credits at https://openrouter.ai/settings/credits, or lower max_tokens / prompt size to fit your remaining balance."
}

Why a small request can fail

OpenRouter charges when a request finishes, so it holds money up front. Its limits page says it estimates each paid request's cost from "the input tokens, plus the completion tokens allowed by max_tokens." If you allowed 16,000 tokens of reply, it prices all 16,000, even if the model would have written 300.

That makes the numbers in the message easy to read. "Requested up to 16000" is your max_tokens. "Can only afford 15154" is how many reply tokens your remaining credit covers at this model's price. The gap between them is the whole problem.

When you never set max_tokens

This catches people whose tools don't send a limit at all. The limits page says a fixed cap is used "when max_tokens is not set," and an oh-my-pi bug from August 2026 shows what that means in practice. A user running GPT-5.6 Luna saw "You requested up to 65536 tokens", but the tool hadn't asked for 65,536. The project's investigation found that OpenRouter "runs its pre-flight credit check against the model's default max output (65536 for Luna)" when no cap is sent. The same account, at the same moment, ran a title-generator call with max_tokens set to 1024 without trouble.

Three places the limit can come from

• The account balance. The limits page adds that a negative balance can produce 402s "including for free models."
• A limit on the API key. Keys can carry their own credit cap. With plenty of money in the account, this is the first suspect; GET /api/v1/key reports limit_remaining.
• The in-flight budget. OpenRouter caps what can be reserved across requests running at once, at a fraction of your balance up to a fixed ceiling. That shows up as limit_source openrouter_in_flight_budget with a different message, and it clears on its own after the Retry-After wait.

The fix

Set max_tokens explicitly and keep it under the "can only afford" figure (in a tool, look for a "max output tokens" or similar setting). If you'd rather not cap replies, add credits until the reservation fits.

Don't just retry. OpenRouter says for the single-request case that "retrying will not help," and none of the major SDKs retries a 402 on its own anyway. A tool that swallows the error in a retry loop will keep failing until one of those numbers moves.