Guide · AI Receptionists
Per call, per minute, or flat: your average call length decides the billing model
The same 200 calls cost wildly different amounts under each model — short calls reward per-minute, long calls reward per-call or flat, and the crossover is shorter than most buyers guess.
The billing model that wins is the one that matches your average call length, and the math is unforgiving about it. Take 200 inbound calls a month. Under Smith.ai’s per-call ladder, the cheapest published route to that month is its self-service Pro — 150 calls/mo rung — $270 for 150 calls, then $2.30 for each of the other 50 — which is $385 whether every call runs 90 seconds or nine minutes; the meter doesn’t care. Under Bland’s all-in per-minute rate of $0.14/min, the same 200 calls cost $56 if they average two minutes but $168 if they average six. And under a flat unlimited plan like NextPhone Pro at $199/mo, the answer is $199 no matter what. There is no universally cheapest model — only the cheapest model for your call length.
Smith.ai rebuilt that ladder on 2026-09-09, and it is worth saying what moved, because this guide used to model the vendor at a flat $2.40 per call: that rate is gone from the page entirely, replaced by six rungs stepping from $3.00 on a new free tier down to $2.10 on a sales-led Enterprise plan. The model didn’t change — Smith.ai still bills the call, not the minute — but every number attached to it did.
The same workload, three meters
Fix the volume at 200 calls and vary only the average duration. Here is what each billing model charges, using each vendor’s honest end-to-end rate where one exists.
| Vendor | Model | Rate | 200 short calls (2 min) | 200 long calls (6 min) |
|---|---|---|---|---|
| Vapi | per-minute | $0.05/min | $20 | $60 |
| Retell AI | per-minute | $0.07/min | $28 | $84 |
| Synthflow | per-minute | $0.09/min | $36 | $108 |
| Bland | per-minute | $0.14/min | $56 | $168 |
| CallSprout Sage | per-minute | $22.95 per user seat/mo + $0.39/min | $178.95* | $490.95* |
| Upfirst (Pro) | per-call | $159.95/mo, 300 calls | $159.95 | $159.95 |
| Goodcall | flat | $79/mo unlimited | $79 | $79 |
| NextPhone | flat | $199/mo unlimited | $199 | $199 |
| Smith.ai (Pro — 150 calls/mo, self-service) | per-call | $270/mo for 150 calls, then $2.30 | $385 | $385 |
* One seat, and an upper bound: our cost model can’t express Sage’s free first 30 seconds of every call, so it bills every modelled minute and overstates by exactly $0.195 a call. The real bills are $139.95 and $451.95 — and five seats would add $91.80 a month to both.
The per-minute column slides with duration; the per-call and flat columns are flat horizontal lines. That is the entire decision in one table: per-minute bills the axis that moves, flat and per-call bill the axis that doesn’t.
Where the lines cross
The interesting number is the call length where one model overtakes another. Against Bland’s $0.14/min — the one per-minute rate in this dataset that’s explicitly all-in, with model, speech, and telephony included — Upfirst’s $0.70/call floor rate (its Scale tier overage) breaks even at exactly five minutes per call ($0.70 ÷ $0.14). Below five minutes, paying per minute is cheaper; above it, paying per call wins.
Smith.ai’s crossover is no longer a single number, because its per-call rate is no longer a single number. Its cheapest marginal call — $2.10, on the sales-led Enterprise 500 rung — breaks even against Bland at 15 minutes; its dearest, the $3.00 charged on the free tier, at just over 21. The rung that actually governs a 200-call month, the $2.30 on Pro — 150, crosses at about 16 and a half minutes. Take the whole bill rather than a marginal call and the picture is friendlier but still long: $385 across 200 calls is $1.93 an answered call, which catches Bland’s meter at just under 14 minutes. Wherever you land in that 14-to-21-minute band, the conclusion is the one this guide has always drawn — Smith.ai’s per-call model only makes sense for genuinely long, high-value conversations like legal intake, the market it serves.
CallSprout Sage crosses these lines from the opposite direction, and it is the clearest illustration of the axis in the whole dataset. It meters per minute, but only after the first 30 seconds of every call, billed by the second — so a per-call discount rides on a per-minute meter. A one-minute call costs $0.195, because half of it is free. A two-minute call costs $0.585, an effective $0.29/min. A six-minute call costs $2.145, an effective $0.36/min. Sage’s effective rate therefore rises toward its $0.39 headline as calls get longer, which is exactly the opposite of what a per-call plan does — there, the effective per-minute rate falls the longer people talk. Short callers should want Sage’s meter; long callers should want somebody’s bundle.
So the rule of thumb writes itself. A nail salon or a dental office fielding 90-second “are you open / can I book Tuesday” calls should run per-minute and pocket the difference. A law firm or contractor running 8–12 minute consultations is overpaying on per-minute and should look at per-call or flat.
The per-minute floor has fine print
The cheap end of the table needs a warning label. Vapi’s $0.05/min that multiplies out to $20 is the platform fee only — the language model, speech-to-text, and text-to-speech are passed through on top, so your real bill is higher. Synthflow’s $0.09/min is the voice engine alone, with LLM ($0.02–$0.05/min) and telephony additive. Retell’s $0.07/min is the published floor of a $0.07–$0.31/min range. Only Bland’s $0.14/min is a true all-in figure you can take at face value, which is why it’s the per-minute number worth anchoring on.
Sage’s fine print is a different shape, and larger than any of those. Its $22.95 buys one user seat, not a business — the only per-seat price in this 25-vendor index — so a five-person office starts at $114.75/mo before a single minute is metered, and Sage is one included feature of a cloud phone system rather than a receptionist subscription sold on its own. Nothing is bundled: zero Sage minutes, so the $0.39 runs from the first call. A second meter runs alongside it, AI Transcription at $0.0198/min on every plan (voicemail transcription is free), which would add $7.92 to the two-minute month and $23.76 to the six-minute one — the page doesn’t say whether a Sage-answered call also incurs it, so it isn’t folded into the table above. Read together, those make Sage the sharpest sticker-versus-bill gap in the index: $22.95 is the lowest paid base price in it, below Upfirst and AIRA at $24.95 and ReceptionHQ at $25, with only $0 free tiers beneath it, yet at 200 calls averaging three minutes one seat models at about $257 — fifteenth of the 21 vendors we can price at that workload.
When flat is the smart default
Flat unlimited stops being about price and starts being about predictability. Goodcall’s Starter at $79/mo carries unlimited minutes — 200 calls or 2,000, the line item doesn’t move (its only cap is 100 unique customers, then $0.50 each). NextPhone Pro at $199/mo is unlimited inbound with no per-minute fees at all. Neither is the cheapest at low volume — Goodcall’s $79 loses to Vapi’s $20 and Bland’s $56 on short calls — but both win the moment your volume or call length becomes unpredictable. Flat is the model you buy to stop thinking about the meter, and it is the exact opposite of a plan like Sage’s, which bundles nothing and meters everything.
How to read this
Don’t ask which billing model is cheapest; ask how long your average call runs. Under two minutes, per-minute wins — Bland at $56, the raw platforms lower still if you’ll assemble the stack. Sage’s free first 30 seconds bites hardest at those lengths too — a two-minute call bills as 90 seconds — but one seat still lands at $139.95 for the month once that waiver is counted, so the discount narrows the gap without closing it. Five minutes is the crossover where per-call (Upfirst) catches up to all-in per-minute. Past that, or whenever volume is erratic, flat wins — Goodcall at $79 or NextPhone at $199 — and Smith.ai’s $385 per-call month, which needs calls of roughly 14 minutes before it beats an all-in meter, is reserved for the long, high-stakes calls where a flat per-conversation price is a feature, not a penalty. Measure your own call length first; the model follows from it.