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