In this guide
  1. What changed on October 30, 2026
  2. Alerts vs hard limits at a glance
  3. OpenAI API: turn on “Enforce a hard limit”
  4. Anthropic API: organization and workspace limits
  5. Google Gemini API: project spend caps
  6. Zapier: the per-run task limit and pay-per-task
  7. Make: steps per agent call and credits
  8. n8n: Max Iterations, then cap the model key
  9. Checklist: before an agent goes live
  10. FAQ

TL;DR: An alert is not a cap. Turn on a hard limit at each layer that bills you: the model provider (OpenAI, Anthropic, Google’s Gemini API) and the automation tool (n8n, Make, Zapier). Give each agent its own project or workspace so one runaway agent can’t use up everyone’s budget. OpenAI and Google both say enforcement can lag a little, so set caps below what you can afford to lose.

Disclosure: The Make link on this page is an affiliate link: we may earn a commission if you sign up through it, at no extra cost to you. It doesn’t change our advice. There are no other affiliate links on this page. See our affiliate disclosure. Method: this is a sourced explainer. Every setting comes from the vendor’s own documentation, read on October 18, 2026 (Google Cloud spend caps: October 31, 2026). We haven’t tested these settings ourselves, so nothing here is a test result.

What changed on October 30, 2026

Google Cloud now has real spend caps. On October 30, Google announced that spend caps in Cloud Billing budgets are generally available (its billing release notes date general availability to October 13). A spend cap budget pauses new usage of one service in one project once estimated spend passes your monthly amount. Eligible services are the Gemini API, Vertex AI (now called Gemini Enterprise Agent Platform), Cloud Run and Cloud Run functions. We’ve updated the Gemini section below. Our earlier line said a Google Cloud budget can’t stop spending, and that’s no longer true for these services.

Alerts vs hard limits at a glance

Layer Only notifies Actually stops spending
OpenAI API Spend alerts Monthly spend limit with “Enforce a hard limit” on (organization or project)
Anthropic API Spend alerts on a workspace Organization spend limit; workspace spend limits
Gemini API Alerts-only Cloud Billing budgets Project spend cap (experimental); Cloud Billing spend cap budget; billing-account tier cap; Prepay balance at $0
Zapier — Per-run task limit on AI steps (pauses for approval); pay-per-task off, or the 3× cap with it on
Make — Steps per agent call; scenarios stop when credits run out
n8n — Max Iterations on the AI Agent node; Gateway credits stop at a $0 balance (n8n Cloud); otherwise cap the provider key

OpenAI API: turn on “Enforce a hard limit”

From OpenAI’s spend limits guide:

  1. Organization: go to Organization limits. In Spend, select Edit spend limit, enter the Monthly spend limit, turn on Enforce a hard limit, and select Save.
  2. Project: go to Project settings > Limits. In Spend, select Edit spend limit, enter the Monthly spend limit, turn on Enforce a hard limit, and select Save.

What to know:

  • Without the toggle, you only get alerts. OpenAI says plainly: “Spend alerts do not enforce a cap.”
  • At the limit, requests return a 429 error with the code organization_spend_limit_exceeded or project_spend_limit_exceeded. The limit resets with the next monthly cycle unless you raise or remove it.
  • “Enforcement is not instantaneous, so recorded spend can slightly exceed the configured amount.”
  • An organization hard limit covers every project; a project hard limit covers only that project. Put each agent in its own project with its own cap.

Anthropic API: organization and workspace limits

From Anthropic’s rate limits and workspaces docs:

  • Organization: in the Claude Console, open the Billing page. In the Spend limits section, click Adjust limit (or Set limit if none is set) and enter an amount. It can’t exceed your usage tier’s own monthly cap.
  • Workspace: on the Workspaces page, select the workspace, then click Spend limits to cap monthly spending and set alerts at thresholds. A workspace limit can’t be higher than the organization’s, and you can’t set limits on the Default Workspace. So create a separate workspace for each agent.
  • When usage reaches a limit you set, requests return HTTP 400. The message begins “You have reached your specified API usage limits” (or “…workspace API usage limits”) and says when access resumes.
  • Separately, each usage tier carries an Anthropic-set monthly spend cap ($500 on Start, $1,000 on Build). Reaching it pauses usage until the start of the next month unless you request more.

Google Gemini API: project spend caps

From Google’s Gemini API billing docs (last updated September 28, 2026):

  • Project cap: in Google AI Studio, open the project’s Spend page and choose Monthly spend cap > Edit spend cap. You need the project editor, owner or admin role.
  • Google labels project caps experimental: “You are subject to overages for around a 10 minute latency period,” and “long-running tasks like batch mode completions and agent sessions may incur overages beyond your project spend cap.”
  • Billing-account tier cap: each paid tier has a monthly cap ($250 on Tier 1, $2,000 on Tier 2). When the account total reaches it, service pauses for every linked project until the 1st of the next month.
  • Prepay balance: on the Prepay plan, when your credit balance hits $0, all API keys in all projects on that billing account stop working (HTTP 402). If you turn on auto-reload, set the monthly auto-charge limit too, or reloads keep refilling the balance. Google notes Prepay is also “subject to overages during the approximately 10 minute billing pipeline latency,” and long-running agents can keep consuming credits past $0.
  • Gemini spend caps (project and billing-account tier) aren’t available for invoiced (offline) Cloud Billing accounts, and neither is Prepay.
  • A regular (alerts-only) Google Cloud Billing budget doesn’t stop spending: Google says it “doesn’t automatically cap” usage. (Google Cloud: Create a budget)
  • Spend cap budget (Google Cloud; generally available October 13, announced October 30, 2026): a different kind of budget that does stop spending, per Google’s spend cap docs. Each one covers one project and one eligible service: the Gemini API, Vertex AI (Gemini Enterprise Agent Platform), Cloud Run or Cloud Run functions. When estimated gross spend passes 100% of the monthly amount, new requests to that service in that project are paused. Requests already running finish and are billed. Alerts go out at 50%, 80% and 100%. The cap lifts on the 1st of the next month, or when you lift it by hand. You can also set caps through the Cloud Billing Budget API, for example on every new sandbox project.
  • Google’s limits on spend caps: enforcement “isn’t instant and any cost overages are billed as normal”; caps can’t cover several projects, services, folders or a whole organization; subscription costs aren’t covered; reseller accounts can’t use them. Google’s blog adds that Provisioned Throughput is never paused.

Zapier: the per-run task limit and pay-per-task

  • Per-run limit on AI steps: an AI by Zapier step has a task limit per run. “The default is 75 tasks, and you can raise it up to 500 tasks.” If a step reaches it during a live run, Zapier pauses the step and asks you to approve before it continues. (Zapier’s October 6 announcement still called the limit “adjustable in future”; if you don’t see the setting, the 75-task default applies.) (Zapier Help: model tier pricing)
  • Account-wide stop: with pay-per-task billing on, Zaps stop when usage reaches 3 times your plan’s task limit. To stop at your plan’s limit instead, the account owner turns pay-per-task off under Billing and usage > Billing settings. (Zapier Help: pay-per-task billing)
  • Model tier: new AI steps on paid plans default to the Premium tier (5 tasks per run plus 5 per tool call), so picking a lower tier makes each run cheaper. Turn on Require approval before running for tools that send, change or delete data. (Add tools to your AI by Zapier step)

Make: steps per agent call and credits

  • Steps per agent call: in the Make AI Agent (New) app’s Run an agent module, this is “the maximum number of times the agent calls the model per request.” Set it so a looping agent stops. (Make AI Agent (New) app)
  • New to Make agents? Our Make AI Agents tutorial walks through building one, including where this setting lives.
  • Credits: when credits run out, “your scenarios won’t continue to run, until credits are added again.” Incoming webhooks queue up to your storage limit. If you want that stop to hold, leave Extra credits auto-purchasing off: it buys 10,000 credits before you run out (Core, Pro and Teams plans). (Make pricing FAQ) See Make’s plans (affiliate link)
  • Your own AI key: if your agent uses your own OpenAI, Anthropic or Gemini connection, model costs are billed by that provider, so cap the key there as well.

n8n: Max Iterations, then cap the model key

  • Max Iterations: the AI Agent node’s option for how many times the model runs to reach an answer. It “defaults to 10.” (n8n docs: Tools Agent)
  • Executions: on Business and Enterprise, n8n’s pricing FAQ says workflows “will continue running without interruption” past your quota, and overage charges may apply. (n8n pricing)
  • Gateway credits (n8n Cloud Starter and Pro): if a node uses Use Gateway credits instead of your own key, n8n bills model usage from a prepaid balance. “When your balance reaches zero, nodes using Gateway credits stop working until you add credit,” though in-flight requests can push it slightly below zero. If you turn on auto top-up, also turn on Monthly auto top-up limit. (n8n docs: Gateway credits, Top up Gateway credits)
  • The real cap is at the provider: when the agent’s chat model uses your own OpenAI, Anthropic or Google Gemini credential, the provider bills you, so set the caps above on that key.

Checklist: before an agent goes live

  1. One provider project (OpenAI, Gemini) or workspace (Anthropic) per agent, each with its own key.
  2. A hard monthly cap on that project or workspace. On OpenAI, check that Enforce a hard limit is on.
  3. Alerts at about 50% and 80% of the cap, so the cap is never your first warning.
  4. In the automation tool: a low Steps per agent call (Make) or Max Iterations (n8n), or keep the 75-task per-run limit (Zapier).
  5. Auto-buying off until you know your usage: Make’s extra-credit auto-purchasing, Zapier’s pay-per-task, and Gemini auto-reload (or a monthly auto-charge limit), and n8n Gateway credits auto top-up (or its monthly limit).
  6. Human approval on any tool that sends, pays or deletes.
  7. Leave headroom: OpenAI and Google both say spending can briefly pass a cap before it takes effect.
  8. On Google Cloud, add a spend cap budget for the Gemini API or Vertex AI in each agent’s project, as well as alerts.

Caps limit the damage from a looping agent, but not from one that acts on bad data. For that, see our verification prompt for agent tool outputs. To compare how the three automation tools bill, see n8n vs Make vs Zapier for AI agents.

FAQ

Does a spend alert stop charges? No. OpenAI says alerts don’t enforce a cap. You have to turn on the hard limit.

Can I end up over my cap? Yes, slightly. OpenAI says enforcement “is not instantaneous,” and Google says Gemini project caps can be passed for about 10 minutes and by long-running agent sessions.

Which layer should I cap? The provider key every time, because that’s where the dollars are. Then add the per-run limits in your automation tool.

Does n8n have a dollar cap? Only for Gateway credits on n8n Cloud, which stop at a $0 balance. With your own keys, use Max Iterations and cap the provider key.

Sources & method

Method. Last checked October 18, 2026; the Google Cloud spend cap material was checked October 31, 2026. Sourced explainer from vendor documentation only; no hands-on testing claimed. Affiliate links: Make only (marked). How we work.