Skip to content

Concurrent streams

1 stream

one thing talking at a time, usage on it unmetered

A stream carries one synthesis at a time, and everything said on it is unmetered. The only number you size is how many you need running at once.

The quiet cap in metered plans

Everywhere, parallel capacity means buying more streams. What differs is what sits on top: a per-character plan also bills every character, and caps how many you may buy. Here nothing is billed by the character. A stream is unlimited and unmetered.

The three mechanics

fig.

Capacity, priced two ways

Fig

1 stream

one thing talking at a time, everything on it unmetered, $150 a month annual

$10

a burst stream-day, when traffic spills past your committed streams

A request that cannot be served is refused fast and retryably, one busy lane never quietly slows the others.

Sizing a fleet

  • Count peak simultaneous work, not monthly minutes. Calls, narration jobs and render batches all draw on the same streams.
  • Let bursts spill instead of sizing for them: the spike days cost $10 a stream-day, with a spend limit you set in the dashboard, and the quiet days cost nothing extra.
  • A first month exists to find the real number, against actual traffic rather than a forecast.

Every term on this page is measurable. Take a key and read the numbers off your own requests.

Get a key and measure it yourself