Skip to main content

OpenAI

Lightweight rate-limit probe for the OpenAI API. OpenUsage issues a single header-only request and parses RPM and TPM limits — no billing data, no token counts.

At a glance​

  • Provider ID — openai
  • Detection — OPENAI_API_KEY environment variable
  • Auth — API key
  • Type — API platform (header-only rate limits)
  • Tracks:
    • RPM and TPM rate limits (limit, remaining, reset)
    • Auth status

Setup​

Auto-detection​

Set OPENAI_API_KEY. OpenUsage registers the provider on next start.

Manual configuration​

{
  "accounts": [
    {
      "id": "openai",
      "provider": "openai",
      "api_key_env": "OPENAI_API_KEY",
      "base_url": "https://api.openai.com",
      "extra": {
        "probe_model": "gpt-4.1-mini"
      }
    }
  ]
}

probe_model defaults to gpt-4.1-mini. Override base_url for proxies or Azure-style gateways.

Data sources & how each metric is computed​

OpenUsage sends one GET https://api.openai.com/v1/models/{probe_model} per poll cycle (default every 30 seconds in daemon mode). The probe model is gpt-4.1-mini unless extra.probe_model is set. The endpoint is read-only, returns a small JSON body that the provider discards, and is not billable.

Request headers:

  • Authorization: Bearer $OPENAI_API_KEY

rpm — requests per minute​

  • Source: response headers
    • x-ratelimit-limit-requests
    • x-ratelimit-remaining-requests
    • x-ratelimit-reset-requests
  • Transform: copied verbatim into Limit / Remaining. Reset is decoded into Resets["rpm"].
  • Window: 1 minute. Scope: the probe model only — different models can have different per-minute limits on the same key.

tpm — tokens per minute​

  • Source: response headers
    • x-ratelimit-limit-tokens
    • x-ratelimit-remaining-tokens
    • x-ratelimit-reset-tokens
  • Transform: same shape as rpm but for tokens.

Auth status​

  • Source: HTTP status code.
  • Transform: 401/403 → auth; 429 → limited (with retry_after from Retry-After if present); otherwise ok.

What's NOT tracked​

  • Spend / cost. OpenAI's API does not expose dollar figures or token usage to API keys. The Usage page on platform.openai.com is a session-cookie surface and is not polled by this provider.
  • Account-wide rate limits. The numbers are scoped to the probe model.

How fresh is the data?​

  • Polled every 30 s by default. One request per poll, no cache.

API endpoints used​

  • GET /v1/models/{probe_model} — header-only probe (default gpt-4.1-mini).

Caveats​

note

OpenAI's API does not expose billing or token-usage data to API keys. OpenUsage cannot show spend for OpenAI; use Codex CLI or OpenRouter to see actual usage data.

  • Rate limits come from response headers; they reflect the probe model's quota, not your account-wide spend.
  • The probe is a single request per poll cycle — negligible cost.

Troubleshooting​

  • Auth failed — verify OPENAI_API_KEY is set and valid; rotate if leaked.
  • No data — the probe model may be unavailable on your tier. Set probe_model to a model your key can access.

Why is there no $ spend?​

OpenAI does not return billing or usage figures on its rate-limit headers, and the Usage and Billing pages are session-cookie surfaces, not API endpoints accessible with a key. Codex (for ChatGPT Pro/Plus accounts) and OpenRouter (when proxying OpenAI) both expose actual usage; either provider gives you a real dollar tile.

Why are my RPM/TPM different from the OpenAI dashboard?​

The numbers come from headers attached to a request for probe_model. Different models share different rate-limit pools on the same account. Set extra.probe_model to the model you actually call most.

  • Codex CLI — OpenAI's coding agent with local session and credit data
  • OpenRouter — proxy with full billing visibility for OpenAI models