> This is a page from the ElevenLabs documentation. For an index of the documentation, fetch https://elevenlabs.io/docs/llms.txt.

# Concurrency estimator

> Estimate the concurrency limit your ElevenAgents, Text to Speech or Speech to Text workload needs from your traffic volume and hours of activity.

The concurrency estimator recommends a concurrency limit from your traffic volume, how long each request lasts and when your traffic happens. Share the result with your account team when you discuss your plan.

#### Voice agents

**Concurrency estimator for voice agents (interactive).** For inbound calls it asks for call volume per day, week or month; the average length; when it happens (24/7, extended hours of 12 hours a day, or a working day of 8 hours); and, unless it's 24/7, whether that's 5 or 7 days a week. Under Advanced it asks how solid the numbers are and how sure you need to be of covering normal ups and downs. For outbound campaigns it asks for the number of contacts, the hours to finish, the share of calls that connect, the average connected call length and whether Batch Calling is used, and under Advanced the time on unanswered attempts, whether and when unanswered contacts are called again, and how sure you need to be of finishing on time.

The average is arrivals per active minute times the average length, with a cushion for how solid the numbers are (1.0, 1.2 or 1.5). The expected peak adds z × √average for normal ups and downs, with z = 1.28, 1.64, 2.33 or 3.09 for 90%, 95%, 99% or 99.9% confidence. The minimum covers the average and the maximum covers the expected peak. The estimate assumes traffic is spread across the active hours: traffic that arrives in large bursts is discussed with your account team. At the minimum, burst covers the ups and downs: ElevenAgents can run up to 3× the limit (at most 300 concurrent calls outside Enterprise), with calls above the limit billed at twice the standard rate. Outbound concurrency is the campaign's total line minutes divided by the calling window. Batch Calling can use up to 50% of workspace concurrency and 70% of an agent's own limit, so the workspace limit is the campaign's lines ÷ 50%.

Example: 2,000 inbound calls a day, 4 minutes each, over a working day of 8 hours, 5 days a week, gives a recommended limit of 20 to 28 concurrent calls. A limit of 20 relies on burst for normal ups and downs; 28 covers them without burst.

These are planning estimates, not a guarantee of capacity. Final limits are confirmed with your ElevenLabs account team.

#### Chat agents

**Concurrency estimator for chat agents (interactive).** It asks for chat volume per day, week or month; the average length; when it happens (24/7, extended hours of 12 hours a day, or a working day of 8 hours); and, unless it's 24/7, whether that's 5 or 7 days a week. Under Advanced it asks how solid the numbers are and how sure you need to be of covering normal ups and downs.

The average is arrivals per active minute times the average length, with a cushion for how solid the numbers are (1.0, 1.2 or 1.5). The expected peak adds z × √average for normal ups and downs, with z = 1.28, 1.64, 2.33 or 3.09 for 90%, 95%, 99% or 99.9% confidence. The minimum covers the average and the maximum covers the expected peak. The estimate assumes traffic is spread across the active hours: traffic that arrives in large bursts is discussed with your account team. Chats aren't queued or covered by burst: a chat that starts while the workspace is at its chat-only limit can't connect. The chat-only limit is separate from the voice limit.

Example: 5,000 chats a day, 8 minutes each, over extended hours of 12 hours a day, 7 days a week, gives a recommended limit of 67 to 81 concurrent chat sessions.

These are planning estimates, not a guarantee of capacity. Final limits are confirmed with your ElevenLabs account team.

#### Text to Speech API

**Concurrency estimator for the Text to Speech API (interactive).** For conversations, such as your own voice agent, it asks for conversation volume per day, week or month; the average length; when it happens (24/7, extended hours of 12 hours a day, or a working day of 8 hours); and, unless it's 24/7, whether that's 5 or 7 days a week. Under Advanced it asks how solid the numbers are and how sure you need to be of covering normal ups and downs, plus the Text to Speech model. For bulk generation it asks for the characters per job, the jobs running at the same time, the time to finish and the model, and under Advanced the headroom for slower generation.

The average is arrivals per active minute times the average length, with a cushion for how solid the numbers are (1.0, 1.2 or 1.5). The expected peak adds z × √average for normal ups and downs, with z = 1.28, 1.64, 2.33 or 3.09 for 90%, 95%, 99% or 99.9% confidence. The minimum covers the average and the maximum covers the expected peak. The estimate assumes traffic is spread across the active hours: traffic that arrives in large bursts is discussed with your account team. Requests above a Text to Speech limit are served at lower priority with a small added latency instead of failing. One request slot serves about 20 live conversations, and plans allow twice the concurrency on Flash v2 and v2.5. Bulk generation is sized by throughput: the characters to generate ÷ (characters per second one request generates × the time available), plus headroom.

Example: 20,000 conversations a day, 5 minutes each, over extended hours of 12 hours a day, 7 days a week, gives a recommended limit of 9 to 10 concurrent Flash requests.

These are planning estimates, not a guarantee of capacity. Final limits are confirmed with your ElevenLabs account team.

#### Speech to Text API

**Concurrency estimator for the Speech to Text API (interactive).** For realtime transcription it asks for session volume per day, week or month; the average length; when it happens (24/7, extended hours of 12 hours a day, or a working day of 8 hours); and, unless it's 24/7, whether that's 5 or 7 days a week. Under Advanced it asks how solid the numbers are and how sure you need to be of covering normal ups and downs. For batch files it asks for files per day, week or month, the average file length and when files arrive, and under Advanced the most files submitted at once, how solid the numbers are and the confidence needed.

The average is arrivals per active minute times the average length, with a cushion for how solid the numbers are (1.0, 1.2 or 1.5). The expected peak adds z × √average for normal ups and downs, with z = 1.28, 1.64, 2.33 or 3.09 for 90%, 95%, 99% or 99.9% confidence. The minimum covers the average and the maximum covers the expected peak. The estimate assumes traffic is spread across the active hours: traffic that arrives in large bursts is discussed with your account team. Realtime requests above the limit are served at lower priority with a small added latency instead of failing. Batch Speech to Text splits files into parts of about 8 minutes, up to 4 per file, each using one slot, and transcribes about 30× faster than realtime. Batch requests above the limit are rejected with a 429 error, so send files from a queue.

Example: 3,000 realtime sessions a day, 6 minutes each, over a working day of 8 hours, 5 days a week, gives a recommended limit of 45 to 56 concurrent realtime sessions.

These are planning estimates, not a guarantee of capacity. Final limits are confirmed with your ElevenLabs account team.

> **Warning**
>
> These are planning estimates, not a guarantee of capacity. They assume your traffic is spread
> across your active hours. If a large share of it arrives at once, talk to your account team before
> you plan a limit around it. Final limits are confirmed with your ElevenLabs account team.

## Next steps

* [Concurrency and priority](/docs/overview/models#concurrency-and-priority) lists the Text to Speech and Speech to Text limits for each plan and explains how requests count toward them.
* [How many ElevenAgents requests can I make?](/docs/help-center/product/eleven-agents/how-many-eleven-agents-requests-can-i-make-and-can-i-increase-it) covers agent limits by plan.
* [Burst pricing](/docs/eleven-agents/guides/burst-pricing) explains how agents handle calls above the limit, and what those calls cost.
* [Call queueing](/docs/eleven-agents/guides/call-queueing) holds callers instead of rejecting them when an agent is at its limit.
* [Chat mode](/docs/eleven-agents/guides/chat-mode#concurrency-benefits) describes the chat-only limit, which is separate from voice calls.
* [Batch calls](/docs/eleven-agents/phone-numbers/batch-calls#concurrency) explains how much of your concurrency a batch uses.
* [Usage analytics](/docs/overview/administration/usage-analytics) shows your concurrent requests over time, which helps you measure your traffic.