Guide · AI Receptionists
The cheapest developer voice-AI platform per minute isn't the one with the lowest headline rate
Vapi's $0.05 and Retell's $0.055 are platform-only rates; once you add the LLM and telephony they pass through, Bland's all-in $0.11–$0.14 stops looking expensive.
If you read the pricing pages straight, Vapi wins this category at $0.05/min and Retell is next at $0.055/min. Both numbers are real, and both are misleading, because neither is a price you can actually run a call on. Vapi’s $0.05 is a platform fee; the language model, speech-to-text, and text-to-speech are passed through at cost on top. Retell’s $0.055 is voice infra only — its own page quotes an all-in range of $0.07–$0.31/min once you add an LLM and the $0.015/min US telephony. The honest floor in this market is not $0.05. It’s closer to $0.07, and the only vendor that quotes a single number you can take at face value is Bland, at $0.11–$0.14/min all-in.
What each headline rate actually covers
The four developer platforms here bill in fundamentally different ways, and the difference is entirely in what the rate bundles.
| Platform | Headline rate | What it includes | Honest floor |
|---|---|---|---|
| Vapi | $0.05/min | Platform fee only — LLM/STT/TTS passed through at cost | $0.05 + model stack (unquantified on page) |
| Retell AI | $0.055/min | Voice infra only | $0.07/min all-in (infra $0.055 + telephony $0.015) |
| Synthflow | $0.09/min | Voice engine only — LLM and telephony additive | $0.11/min (engine + $0.02 LLM floor) + telephony |
| Bland | $0.11–$0.14/min | All-in: LLM + STT + TTS + telephony | $0.11/min (Scale) to $0.14/min (Start) |
Vapi is the trap. The $0.05/min is genuinely the lowest platform fee on the table, and if you bring your own keys it stays low — but the dataset’s own provenance is explicit that “true end cost depends on chosen models,” and the page does not quantify the pass-through. You cannot price a Vapi minute from the pricing page alone. What you can price is the HIPAA add-on, which is a flat $2,000/mo — a number that inverts the entire “cheapest” ranking the moment compliance enters the picture.
Retell is more transparent: $0.055 infra plus $0.015 telephony gives a $0.07 floor, but the published ceiling is $0.31/min depending on which LLM you wire up. That’s a 4.4x spread inside one vendor, and the floor only holds if you run the cheapest model.
Synthflow’s $0.09/min voice engine is the highest headline of the four, and it’s still not all-in — the LLM adds $0.02–$0.05/min and telephony is separate again. Stack the cheapest case and you’re at roughly $0.11/min before telephony, which lands it in the same neighborhood as Bland’s all-in rate.
The all-in comparison flips the order
Once you normalize to what actually leaves your account per minute, the headline ranking falls apart:
- Retell, cheapest LLM: $0.07/min all-in.
- Bland Scale: $0.11/min all-in (with a $499/mo platform fee).
- Synthflow, cheapest LLM: ~$0.11/min plus telephony.
- Bland Start: $0.14/min all-in, $0/mo platform fee.
- Vapi: $0.05 platform fee plus an unquantified model stack — possibly cheapest, possibly not, unknowable from the page.
Bland’s structure rewards volume through the platform fee rather than the per-minute rate: Start is $0.14/min at $0/mo, Build drops to $0.12/min for a $299/mo fee, and Scale reaches $0.11/min for $499/mo. The crossover matters. At low volume, Start’s zero platform fee wins; at high volume, Scale’s $0.11/min pays back the $499. The point where Build’s $299 beats Start is around 15,000 minutes a month ($299 ÷ the $0.02/min saving), and Scale’s extra $200 over Build pays back near 20,000 minutes ($200 ÷ $0.01). Below those thresholds, the free Start tier at $0.14/min is the rational Bland choice.
So which is actually cheapest?
For a buyer who wants the lowest defensible per-minute number with no assembly and no surprise line items, Bland is the answer — $0.14/min on the free tier, dropping to $0.11/min at scale, with nothing passed through. Its provenance is the cleanest in the dataset: a single all-in talk rate, HIPAA-eligible with a signed BAA, no model wiring required.
Retell can beat it — $0.07/min — but only at the floor of a $0.07–$0.31 range, and only if you accept the cheapest LLM and US-only $0.015 telephony. Vapi’s $0.05 is the lowest platform fee and the highest uncertainty: cheap if you optimize your own stack, opaque if you don’t, and $2,000/mo more if you need HIPAA. Synthflow’s $0.09 engine is the weakest headline once the additive LLM and telephony are counted.
The rule for this category: a per-minute platform’s headline rate tells you what the vendor charges for its slice, not what the call costs. Bland prices the whole call; everyone else prices a part of it and lets you discover the rest. If you want a number you can put in a budget today, it’s Bland’s $0.11–$0.14. If you’re willing to engineer and monitor a model stack, Retell’s $0.07 floor — or Vapi’s unknowable-but-low platform fee — can go lower.