OpenAI spending limits: the hard limit most teams never turn on

Last checked: 2026-11-21 ยท Facts follow OpenAI's official documentation, linked throughout.

Soft limits email you. Hard limits stop.

OpenAI has two different mechanisms that are easy to confuse. A budget / soft limit sends an alert when you cross a threshold โ€” spending continues. An enforceable spending limit rejects requests (HTTP 429) once the cap is hit. In 2026 OpenAI made the enforceable limit available at organization and project level โ€” but it is opt-in and off by default. If you never opened that settings page, you do not have a hard limit.

How to turn it on

  1. Open your organization's limits settings (requires an owner/admin role).
  2. Set a monthly hard limit for the organization โ€” this is your outer wall.
  3. Repeat per project: project-level limits stop one runaway project from eating the org budget. See OpenAI's rate & usage limits documentation.
  4. Set alert thresholds well below the cap (50% / 80%) and route them somewhere you look daily.
  5. Use separate API keys per project and per automation, so a limit โ€” or a disable โ€” can target exactly one workload.

The holes that remain

Monthly โ‰  nightly. The cap is monthly. A loop at 3am can spend the whole month before the cap notices a problem โ€” it only stops at the full monthly amount.

Enforcement isn't instant. Usage reporting lags, so spend can overshoot between checks.

It only covers OpenAI. Your Anthropic, OpenRouter, and coding-agent spend lives behind different switches, in different dashboards.

That gap โ€” daily, cross-provider, agent-aware โ€” is what the Hard-Limit Audit checks and the Guard Kit watches.

Run the free Hard-Limit Audit