LLM API providers limit how much you can spend by usage tier. As of 19 October 2026, Anthropic lists monthly caps of $500, $1,000 and $200,000 on its Start, Build and Scale tiers, and pauses access until the next month at the cap. Google lists spend limits per 10 minute window. Set your own lower cap first.
Anthropic: monthly caps
Anthropic's rate limit documentation lists a monthly spend cap by tier:
| Tier | Monthly cap |
|---|---|
| Start | $500 |
| Build | $1,000 |
| Scale | $200,000 |
The Custom tier has no cap and limits are arranged with the account team. Organizations are placed on tiers automatically based on usage history and standing.
What the cap looks like in production
When you reach the tier cap, API usage pauses until 00:00 UTC on the first day of the next month unless you request a higher limit. Requests return HTTP 429 with error type rate_limit_error and an error code of enforced_spend_limit_reached. Unlike a normal rate limit, the response has no retry-after header, and the page says retries, including automatic SDK retries, fail until access resumes.
Your code should treat that case as an outage, not a retry. Read the error code and alert a human.
Setting your own lower cap
You can set a spend limit below the tier cap on the Billing page of the console. When usage reaches a limit you set, requests return HTTP 400 with invalid_request_error, with a message that says when access resumes. You can also set limits per workspace, which helps you stop one experiment from consuming the budget of a production service.
Google: spend limits per 10 minutes
Google's rate limit page lists spend limits over a rolling 10 minute window, by tier: Tier 1 $10, Tier 2 $50, Tier 3 $200. Hitting one returns 429 RESOURCE_EXHAUSTED. Tier qualification, as listed on that date, goes from linking an active billing account (Tier 1), to paying $100 plus 3 days since the first payment (Tier 2), to paying $1,000 plus 30 days (Tier 3). Upgrades are automatic.
A budget checklist
- Set a console limit at the amount you are willing to lose in a bad month, not the tier cap.
- Use separate workspaces or projects for production, experiments and CI.
- Alert at 50 and 80 percent of the limit using your own usage export, since a provider's pause arrives with no warning in the response.
- Decide in advance what your product does when calls start failing: queue, degrade or show an error.
For agent systems that can loop, a hard cap is the last line of defense. See AI agent bills and how they grow.
Quick check
Why did my retries not work after a 429?
If the error code is enforced_spend_limit_reached, you hit a spend cap. Retrying cannot help until the cap resets or is raised.
Can I raise the tier cap myself?
Anthropic documents a request tier increase flow in the console, and says support can raise limits for urgent needs.
Sources
Last reviewed: 19 October 2026.
