OpenAI adds hard spend caps: your API now returns 429 at your limit
Set the ceiling per org or per project; past it, calls fail with insufficient_quota. Alerts warn first — the runaway-agent kill-switch builders wanted.
Copy markdown
The cap, in one line
Set a hard monthly spend limit on your whole org or a single project, and once tracked spend hits it, affected requests fail with a 429 and an `insufficient_quota` code until you raise the ceiling or the month resets. Enforcement isn't instant, so recorded spend can slightly overshoot the number you set.
Alerts fire before the wall
Spend alerts stay active alongside a hard cap, so a warning can land before traffic actually stops. Set the alert below your limit and you get a heads-up instead of a silent outage in the middle of a run.
Per-project caps fence in your agents
Because a limit can be scoped to one project, a looping or runaway agent burns through its own budget and 429s itself without draining the org account. After a summer of rogue-agent incidents, that blast-radius control is the guardrail small teams actually needed.
It's rolling out to every API account
OpenAI says hard spend limits are reaching all API Platform accounts, not just enterprise tiers. Set one in your billing settings before your next unattended or overnight job — it's the cheapest insurance against a five-figure surprise.