Throughput tiers
A tier is a workspace’s requests a minute and calls at once. It rises with spend.
What a tier is
Every workspace has a throughput tier. It sets two limits, checked before credits: requests a minute, as a token bucket, and calls in flight at once. Both models share the same tier. The current budget is in the b11-ratelimit-requests-* headers on every response; see errors.
The tiers
| Tier | Unlocked by | Requests a minute | Calls at once |
|---|---|---|---|
| Tier 1 | free trial | 20 | 2 |
| Tier 2 | card added | 60 | 4 |
| Tier 5 | $100 lifetime spend | 300 | 10 |
| Tier 11 | Enterprise commit | 2,000 | 50 |
Every tier gets the same models. Only Enterprise goes to eleven.
How a workspace moves
- A new workspace on the Free trial is tier 1: 20 requests a minute, 2 at once.
- Adding a card moves it to Pay as you go, tier 2: 60 requests a minute, 4 at once. From there it rises with lifetime spend, up to tier 5: 300 requests a minute, 10 at once. The unlock column above says what each step takes.
- An Enterprise commit is tier 11: 2,000 requests a minute, 50 at once. See Enterprise.
Tier changes are automatic and recorded; the app shows the current tier. A tier is never lowered for spending less. Support can set a tier by hand in special cases.
At the limit
A call over either limit is refused with 429, rate_limited or concurrency_limited, and a retry-after header in whole seconds. Refusals are never charged. Spread calls out, keep a few in flight at most on the lower tiers, and back off on 429. If you hit the limit often, the next tier is a card or a commit away.
Next: Credits and billing