capability
verified July 2026
Burst seats: capacity priced by the day.
Every voice fleet has two sizes: the ordinary day and the spike. Committed seats price the first; burst seats price the second — $10 a seat-day, only on the days the spike is real, with a cap you set.
01
The mechanics
- Traffic past your committed seats spills over automatically — callers hear an answer, not a busy signal.
- Each burst seat bills $10 for that day only; a quiet day bills nothing.
- The dashboard cap is the control: set the maximum burst seat-days you will accept, and past cap the API refuses fast with a retryable 503 instead of surprising finance.
02
The break-even, stated
Burst vs committed — the crossing
Fig
15 days
spike days a month where burst spend matches a committed seat
$10 × days
the whole burst formula — nothing else moves
A seat that spikes three days a month costs $30 on burst against $150 committed. Past fifteen spike days, the honest move is another committed seat — the dashboard cap makes that conversation happen on numbers.
Table 1 — One overflow seat, priced both ways
| Spike days / mo | On burst | A committed seat |
|---|---|---|
| 3 | $30 | $150 |
| 10 | $100 | $150 |
| 15 | $150 | $150 · break-even |
| 22 | $220 | $150 |
03
Where burst earns its keep
The workloads that spike by calendar, not by trend: the 8 AM Monday wave on a front desk, the 9 AM reminder window, the end-of-quarter dialer push, catastrophe morning on a claims desk, peak season on order status. Each of those pages does its own arithmetic; this page is the mechanism they all lean on.
04
Notes — an engineer's checklist
01Do I have to enable burst?
It is a dashboard cap: set it to zero and traffic past your seats gets the retryable 503 instead of spilling. Set it high for storm season, low for the quiet quarter — the knob is yours.
02Is a burst seat slower than a committed seat?
No — a burst seat is the same serving path with the same measured figures. Burst is a billing state, not a serving tier.
03How do I know how many burst days I used?
Burst activations are visible in the dashboard, and each workload’s traffic remains auditable on its own key at GET /v1/usage — so the spike that caused the spend is identifiable, not a mystery line item.
See also — related sheets
glossary
1 seat
Concurrent seats
A seat is one stream of speech, unmetered. Why capacity — not characters — is the unit here, and how metered plans cap seats and bill for everything said on them.
use case
2,000
TTS for insurance claims calls
Claims desks are quiet until catastrophe day. Sizing voice capacity for the storm with burst seats, long structured turns, and one cloned voice in 23 languages.
use case
2,000 min
TTS for appointment reminders and notifications
Reminder traffic is bursty and short-form. Where a meter genuinely wins, this page says so — and where burst seats beat both, it shows the arithmetic.
capability
3 tiers
Serving that degrades before it fails
Redundant serving behind one endpoint, and a retryable error your code can act on instead of a hung request.
A key, one seat, your own script — nothing on it counted while you build.
Request access — run your own script