Every limit, in one place
What is capped
| Limit | Value | On breach |
|---|---|---|
| Sends per minute | per plan, 1 to 300 | 429, Retry-After: 60 |
| Sends per month | per plan, 3,000 to 3,000,000 | 429, Retry-After to next UTC midnight |
| Recipients per send | 100 across To, Cc and Bcc | 400 Too many recipients |
| Recipients before a warning | 25 | 202 with a warning field |
| Attachment, each | 10 MB of file content | 400 Attachment too large |
| Attachments, total | 20 MB | 400 Attachments too large |
| Request body | 30 MB of JSON | 413, with nginx at 32 MB behind it |
| Scheduling horizon | 30 days | 400 Invalid send_at |
| Domain warm-up | a daily cap that rises | 429 |
| Requests per minute | 600, every endpoint | 429, Retry-After: 60 |
| Requests in flight at once | 16 | 429, Retry-After: 1 |
The per-plan numbers are in Rate limits, and
your own are on GET /v1/usage rather than a table you have to match
yourself to.
The general caps
Until October 2026 these did not exist: only sends were limited, and this page said so. A polling loop was load we carried with nothing in front of it, and the honest thing to do about that was to put something in front of it rather than keep describing the hole.
600 requests per minute, counted across every endpoint, sends included. It is far above anything an integration does on purpose. The largest send cap on any plan is 300 a minute, and a read-heavy client that stays under 10 requests a second will never see this. It is a ceiling on an accident, not a budget to manage against.
{
"status": 429,
"title": "Too many requests",
"detail": "This account has made more than 600 requests in the last minute."
}
16 requests in flight at once. Running more in parallel does not make the per-minute budget arrive faster, and past this point the extra ones are refused rather than queued.
{
"status": 429,
"title": "Too many concurrent requests",
"detail": "This account already has 16 requests in flight."
}
The two are reported separately because the fixes differ: widen your polling interval for the first, lower your worker count for the second. Backing off on the wrong one does not help.
What is still not capped
There is no per-endpoint metering. The 600 is one pool across all of them; no single endpoint has a tighter cap of its own, and no endpoint is exempt.
There is no per-key limit. Caps are per workspace, the general ones included. Issuing a second key does not get you a second allowance, and spreading sends across keys to go faster reads as evasion from our side while achieving nothing.
Which limit you hit tells you what to do
| You see | It is | Do |
|---|---|---|
429 Rate cap exceeded |
the per-minute SEND cap | back off, then continue |
429 Too many requests |
the per-minute request cap | back off, then continue |
429 Too many concurrent requests |
too many in parallel | lower your worker count |
429 with a long Retry-After |
the monthly allowance | buy units or upgrade; waiting is until midnight UTC |
429 and your reputation is throttled |
your bounce rate lowered the cap | fix the list, not the retry |
400 Too many recipients |
one send is too wide | use batch |
413 |
the body, not the attachments | see size limits |
See Handling a 429 for the retry shape, and Watching your usage for seeing it coming.