Skip to content
August 2026

Concurrency Limits: The Meter Voice Platforms Skip

For a product sold by the minute, the limit that actually binds is calls in parallel, and buying more of it does not move the per-minute rate.

Ask a voice platform what it costs and you get a per-minute rate. Ask it what happens when eleven people call at once and you usually get nothing at all.

That second number is the one that decides whether the product works on your busiest morning, and on at least one platform in this directory it is the only thing the pricing tiers actually differ on.

ElevenLabs: you pay double for a worse call

Exceed your concurrency limit on ElevenLabs Agents and the calls do not fail. They burst, up to three times the plan's limit, at double the rate.

The detail that matters is what else happens. Burst calls are deprioritised for speech processing and may run at higher latency, which in a voice agent is precisely the failure mode that makes a caller hang up. So the overflow costs twice as much and sounds worse, and past the burst ceiling calls are rejected outright.

Read the tier ladder with that in mind and it changes shape. It is a concurrency ladder with seats and voice clones attached, and somebody buying a higher tier expecting the unit cost to fall has misread what they are paying for.

Ultravox: the cap is the product

Ultravox publishes five cents a minute including synthesis, for up to five concurrent calls. That is on the homepage, in nine words, and it is unusually honest for this category.

The paid tier is not a cheaper rate. It is a concurrency unlock: pay-as-you-go caps you at five simultaneous calls and Pro removes the cap, at the same per-minute price. You are buying the ceiling, not the minutes.

This is the cleanest example on the site of a price attached to a limit rather than to usage, and it is invisible to any comparison built on per-minute rates, because the per-minute rate does not move.

The arithmetic Regal publishes and nobody else does

Regal's concurrency documentation contains the single most useful calculation in this category, and it is about humans rather than software.

Fifty concurrent AI calls, a five-minute average call, a forty percent transfer rate and a thirty-minute human close require about 120 licensed human agents to absorb the handoffs.

That number reframes the whole purchase. Concurrency is not a capacity setting you turn up because you can afford the minutes. It is a commitment to staffing on the other side of the transfer, and buying more AI throughput than you have people to receive is how a successful pilot becomes an abandoned queue.

What to ask before you sign

  • What is the concurrency limit on the tier I am buying, and is it per account or per agent?
  • What happens at the limit: queue, reject, or burst? If burst, at what rate and with what latency?
  • Is concurrency what separates the tiers, or does the per-minute rate change too? If the rate is flat across tiers, you are buying a ceiling.
  • Does an inbound call and its transfer leg count as one concurrent call or two?
  • Run Regal's calculation against your own headcount before sizing the AI side.

The fourth is the one most likely to produce a surprised pause, because on platforms that bill a transfer as a separate leg the answer is two, and your effective concurrency is half what you bought.

Concurrency mechanics read from each vendor's own pricing page and developer documentation in August 2026 and are recorded in the reviews on this site. Regal's 120-agent figure is Regal's own published calculation on its own stated assumptions, quoted as the vendor's arithmetic rather than as a measurement of ours. We have not load-tested any platform.

Questions

What is a concurrency limit on an AI voice agent?
The number of calls the platform will handle at the same time, as opposed to the total minutes you buy. It binds hardest exactly when your phone is busiest, and it is absent from essentially every per-minute comparison published in this category.
What happens when you exceed the concurrent call limit?
It depends on the platform and you should ask. ElevenLabs Agents bursts to three times the limit at double the rate, with burst calls deprioritised for speech processing and therefore higher latency, then rejects calls past the burst ceiling. Others simply refuse the call.
Does paying more reduce the per-minute rate?
Often not. Ultravox's paid tier removes a five-call concurrency cap at an identical per-minute price, so the upgrade buys a ceiling rather than cheaper minutes. Check whether the rate actually changes across tiers before assuming volume discounts exist.
How much concurrency do I actually need?
Less than you think on the AI side and more than you think on the human side. Regal publishes the calculation: 50 concurrent AI calls at a 40% transfer rate and a 30-minute human close needs about 120 licensed agents to absorb the handoffs. Size the humans first.
Does a transfer count as a second concurrent call?
On some platforms yes, and it halves your effective concurrency. Ask explicitly whether an inbound call and its transfer leg count as one or two, because the answer is rarely on the pricing page and it changes the capacity you thought you bought.

Tools mentioned

All tools

Sources

Source interests are labelled. Almost everything published about this subject is written by someone selling into it.

More from the blog

All posts