APIs and integrations
Twilio API Concurrency Limits vs Messages Per Second, and Which One You Hit
In short
Twilio has two separate limits. The REST API limit is about concurrent requests per account: go over it and you get a 429 and nothing is processed, so you back off and retry. Messaging throughput is messages per second per sender: go over it and Twilio queues the messages and sends them at the sender's rate.
Are Twilio's REST API limit and its messages-per-second limit the same thing?
No. They measure different things, sit at different layers and fail in opposite ways. The first one hands the problem back to your code. The second one absorbs it for you and makes messages late.
Twilio says your application can send Programmable Messaging API requests as quickly as it likes, and the queue paces the sending. That sentence is why a message backlog does not by itself cause a REST 429.
| Detail | REST API concurrency | Messaging throughput |
|---|---|---|
| What is counted | Requests in flight for your account | Message segments per second, per sender |
| What happens over the limit | HTTP 429, the request is not processed | Messages queue and go out at the sender's rate |
| Who waits | Your client, by retrying | Twilio, inside its queue |
| Where to look | `Twilio-Concurrent-Requests` response header | Messaging Insights latency report |
What does the REST API concurrency limit measure?
It is a limit on concurrent REST requests for an account, not a count per hour. Twilio says the more concurrent requests or requests per second you send, the more likely a 429 is from some endpoints.
The Twilio-Concurrent-Requests header returns the current number of concurrent requests. The count includes requests that got a 429, and subaccount requests do not roll up to the primary account. The Twilio-Request-Duration header gives processing time inside Twilio. Slow responses keep requests in flight longer, so they can raise concurrency without any change in your send rate. That last link is my reading of the two definitions, not a Twilio statement.
None of the pages I read gives a concurrency number. The error 20429 page lists other causes too, all on Verify: repeated verification starts for one phone number, Verify attempt limits, your own Verify Service Rate Limits, and polling verification status too fast.
What should a client do when it gets a 429?
Twilio says a request that receives a 429 was not processed, so retrying it is safe. Its security FAQ gives the backoff pattern for retryable errors, listed below.
The 20429 page does not mention a Retry-After header or a wait time, so this schedule is the only timing guidance I found. The FAQ also notes that a sudden rise in 429 can signal a compromised credential.
- First retry after 1 second, then 2 seconds, then 4 seconds.
- Keep doubling up to a maximum of 60 to 300 seconds.
- Limit total attempts to 3 to 5, then stop.
- Do not retry other
4xxerrors except in rare transient cases. - Add app-level rate limiting and concurrency controls, or queue requests for retry.
- Use webhooks instead of polling the same resource, and monitor
429spikes.
How does messaging throughput work, and where do messages wait?
Throughput is measured per phone number, called a sender, per second. Twilio counts message segments, abbreviated MPS. Your application sends API requests, and Twilio queues messages and sends them at the rate each sender allows. The docs warn that this keeps you compliant but can delay delivery.
A Messaging Service queues the request before it reaches a sender. The queueing guide says messages are queued for 10 hours (36,000 seconds) by default, and a Validity Period setting controls how long a message stays valid. Use a short one for one-time passcodes, since a code that arrives late is useless.
The queue can overflow. The guide gives error 30001 as an example and says to respond with backoff and retries. It states no maximum queue size.
What throughput does each sender type have?
The numbers depend on the sender type and the country, so check the live pages before you plan around them. These are the figures the two guides state.
Market Throughput is a different model. Capacity is set per country at the parent account level, and it cannot be used for US 10DLC SMS and MMS.
- Short code: typically starting at 100 MPS.
- US toll-free number: a default of 3 MPS, which can be increased.
- Long code outside the US and Canada: 10 MPS. A single number cannot be increased, but you can add more numbers.
- Alphanumeric Sender ID: 10 MPS by default.
- US long code: the queueing guide uses 1 MPS as its example.
How do I size senders so the queue does not make messages late?
Twilio's method is to divide the number of messages by the time you allow in the queue. It suggests a 100% buffer on top, which means planning for twice the calculated rate. Keep one Messaging Service for each use case, each with its own sender pool and validity period.
I would not copy the guide's worked numbers. Its own reference section and its worked example disagree on the promotional case and on the OTP validity window, so I took only the method.
Adding long codes is not a free way to scale. Twilio cautions against adding more long codes to a pool to spread load, and calls multiple numbers for one use case in the US and Canada strongly discouraged because of carrier filtering. It also notes that on a congested network more numbers do not necessarily raise total throughput. For large one-off sends, it says to contact Twilio about adjusted limits.
What do the docs not say?
They give no REST concurrency figure and no statement on whether Twilio queues REST requests. The best-practices page says a 429 means the request was not processed and says nothing more on queueing.
I also did not read the Help Center articles on A2P 10DLC trust scores, toll-free throughput or MMS limits, because those pages did not load for me. WhatsApp, voice and Verify service limits are out of scope here.
What else do people ask?
Is a Twilio 429 safe to retry?
Yes, according to Twilio's best-practices and error 20429 pages. The request was not processed, so retry after backing off.
Does hitting messages-per-second return an error?
Not by default. Twilio queues the message and sends it at the sender's rate, which can delay delivery. A full queue can return an overflow error such as 30001.
Will more phone numbers fix a REST 429?
No. The REST limit is about concurrent requests for the account, not about sender count. Reduce in-flight requests or queue them on your side.
Sources
- Twilio Docs, API best practices, fetched 2026-11-19
- Twilio Docs, Error 20429, Too many requests, fetched 2026-11-19
- Twilio Docs, FAQ about API security, fetched 2026-11-19
- Twilio Docs, Best practices for scaling with Messaging Services, fetched 2026-11-19
- Twilio Docs, Build to scale, queueing and latency, fetched 2026-11-19
- Twilio Docs, Market Throughput overview, fetched 2026-11-19
How was this page made?
I drafted this page with AI assistance from the official documents listed under Sources. This page compares Twilio's own documents, fetched on 19 Nov 2026. I did not send test requests or messages through Twilio for it, so nothing here is a measurement. It does not cover Help Center articles that would not load, 10DLC trust-score rates, WhatsApp, voice or any limit Twilio does not publish. Found a mistake? Tell me.