You exceeded your current quota, please check your plan and billing details

You exceeded your current quota, please check your plan and billing details

The message that made this error famous looks like this, and it is worth saying up front that as of August 2026 it is the legacy form. OpenAI's current error-codes page does not document this string at all.

{
  "error": {
    "message": "You exceeded your current quota, please check your plan and billing details.",
    "type": "insufficient_quota",
    "param": null,
    "code": "insufficient_quota"
  }
}

What replaced it is more useful. OpenAI split the billing 429s into four codes with separate causes and separate fixes, and kept insufficient_quota as the umbrella type. From the current docs, verbatim:

Quota is not rate limiting, and the shared 429 is the whole problem

Two completely unrelated failures share status code 429, which is the single biggest source of confusion here. One is about how fast you are going. The other is about whether you can pay.

Your handler should split on this before it decides anything. A retry loop that cannot distinguish them is a loop that hammers a dead endpoint during an outage of your own making.

import openai

BILLING_CODES = {
    "credit_balance_exhausted",
    "organization_spend_limit_exceeded",
    "project_spend_limit_exceeded",
    "organization_usage_limit_exceeded",
}

try:
    resp = client.responses.create(model="gpt-5.6-terra", input=prompt)
except openai.RateLimitError as err:
    code = (getattr(err, "code", None) or "")
    if code in BILLING_CODES or err.type == "insufficient_quota":
        alert_billing_owner(code)   # a human has to act; retrying cannot help
        raise
    backoff_and_retry(err)          # a real rate limit, safe to retry

The trap: an active subscription and an empty API balance

The most common single cause of this error, by a wide margin, is a developer with a paid ChatGPT subscription who assumes it covers API calls. It does not. ChatGPT and the API platform are separate products on separate billing systems, and money in one is invisible to the other.

• ChatGPT Plus, Pro, Business, and Enterprise pay for chatgpt.com. They add nothing to your API balance.
• The API runs on prepaid credits. No credits, no calls, regardless of what else you pay OpenAI every month.
• They can be different organizations entirely. Your ChatGPT seat may sit in a workspace with no relationship to the org your API key belongs to.
• The failure mode is silent until the first call. The dashboard will happily show a healthy subscription while the API returns credit_balance_exhausted.

The diagnostic takes ten seconds and beats any amount of reasoning about which account you are in:

curl -sS https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.6-luna","input":"ping","max_output_tokens":16}' \
  | jq '.error | {type, code, message}'

Spend limits: the quota you set yourself

Two of the four codes come from limits somebody on your team configured, which means the fix is a settings change rather than a payment. They live in different places and behave differently.

• Hard enforcement is optional. Both limits take a monthly amount, and hard enforcement is a separate toggle. Without it, you get alerts and keep serving traffic.
• Alerts never interrupt service. For most teams this is the correct configuration: you want to know at $500, not to have production stop at $500.
• Limits reset with the monthly cycle. If you are blocked and cannot raise the cap, that is when access returns.
• Project limits isolate blast radius. One project hitting its ceiling leaves the others running, which is the entire reason to put a runaway batch job in its own project.

Usage tiers, and what actually changed

Tiers govern both your monthly usage ceiling and your per-model rate limits, and they graduate automatically as you spend. The table below is current as of August 2026 and differs from what most guides on the internet still show.

• The wait-time requirement is gone. Older documentation and every blog post copying it said Tier 2 needed "$50 paid and 7+ days since your first successful payment". The current qualification column is spend-only.
• There is a Free tier row now, with a $100 monthly usage limit, gated on geography. That is a usage ceiling, not a grant of credits: it does not mean you get $100 of free API calls.
• Tiers move rate limits, not just spend. On gpt-5.6-sol, Tier 1 starts at 500 requests per minute and Tier 5 reaches 15,000. On gpt-5.6-luna, Tier 5 reaches 30,000 RPM and 180M tokens per minute.
• Graduation is automatic as cumulative spend crosses each threshold. There is nothing to apply for below Tier 5.

Fixing it, in order

Questions people ask

Keep reading

Continue through Errors and fixes with Incorrect API key provided: the OpenAI 401 and every reason you get it.