01At a glance
| Critical criterion | Synthflow | Retell AI | Vapi | Winner |
|---|---|---|---|---|
| Advertised price | $0.08/min | $0.07/min | $0.05/min | Vapi on paper |
| Realistic all-in cost | $0.15 to 0.24/min | $0.125 to 0.25/min | $0.15 to 0.40/min | Retell standard stack |
| Included concurrency (PAYG) | 5 calls (+$20/slot) | 20 calls (+$8/slot) | 10 calls (+$10/line) | Retell |
| HIPAA | Enterprise only | Enterprise only (BAA) | +$1,000/mo add-on | None at entry level |
| Build model | No-code, agency white-label | Visual flows + API | API-first, bring your own stack | Depends on your team |
02The pricing trap, quantified
Voice AI pricing has a structural quirk. The advertised number covers orchestration only. A live call also burns speech-to-text, a language model, text-to-speech, and telephony, and each platform splits those costs differently.
Vapi is the sharpest example. The pricing page says $0.05 per minute. That is the hosting fee. Add transcription (about $0.01/min), the LLM ($0.02 to 0.20/min depending on model), and voice generation (about $0.04/min with ElevenLabs) and the cheapest viable production stack lands near $0.15/min. A premium stack with GPT-4o and multilingual voices reaches $0.35 to 0.40/min. Eight times the sticker.
Retell bundles more into its base. A standard setup with platform voices and a mainstream LLM runs around $0.125/min; swap in ElevenLabs voices and Claude and you sit at $0.19 to 0.25. Synthflow charges $0.08 to 0.09 for its voice engine, then $0.02 to 0.04/min for the LLM and $0.02/min for managed telephony, for an effective $0.15 to 0.24/min. Bringing your own Twilio account removes the telephony line entirely.
The 10,000-minute benchmark
Our reference scenario: a business handling 10,000 inbound call-minutes a month. Roughly what a busy two-location clinic or a mid-size home services company generates.
| Platform | Cheapest viable stack | Premium stack | Hidden line items |
|---|---|---|---|
| Synthflow | ~$1,500/mo | ~$2,400/mo | Concurrency slots $20/mo each above 5 included |
| Retell AI | ~$1,250/mo | ~$2,500/mo | Concurrency $8/slot above 20; HIPAA locked behind Enterprise |
| Vapi | ~$1,500/mo | ~$4,000/mo | HIPAA/zero-retention +$1,000/mo; every model billed separately |
03Matrix highlights
Economics and hidden limits (points 1 to 4)
- Billing unit: all three bill per minute, but Vapi unbundles every component while Synthflow and Retell partially bundle. Unbundling is cheaper for engineers who optimize, and more expensive for everyone else.
- Entry cost: all three start at $0 pay-as-you-go. Retell includes $10 in credits. Real entry cost is your first month of testing: budget $50 to 150.
- Concurrency caps (the real rate limit): Synthflow 5, Vapi 10, Retell 20 on entry plans. Extra capacity: $20, $10, and $8 per slot respectively. These caps are the voice equivalent of the 429 errors we documented in the Make vs n8n audit.
- Enterprise floors: Retell's managed tier starts near $8,000/mo; Synthflow enterprise contracts start around $30,000/year. Vapi enterprise is custom [UNVERIFIED].
AI engine (points 5 to 7)
- LLM agnosticism: Vapi wins. Bring any model, any STT, any TTS, billed at cost. Retell offers a curated list including GPT and Claude families. Synthflow offers tiered model choices at fixed per-minute add-ons, which is simpler and slightly opaque.
- Knowledge and memory: all three support knowledge bases for grounding answers. Depth varies and vendor docs are thin here. Lab testing required [UNVERIFIED].
- Multi-agent and transfers: all three handle human handoff and call transfer. Vapi exposes the most control via API; Synthflow packages it into no-code flows.
Infrastructure and compliance (points 11 to 13)
- Hosting: all three are cloud-only. No self-hosted option exists in this category yet. Whoever ships one first changes the market.
- Certifications: Synthflow lists SOC 2, GDPR, and ISO 27001 on pay-as-you-go, with HIPAA gated to Enterprise. Retell is SOC 2 certified and HIPAA-ready via Enterprise BAA only. Vapi sells HIPAA and zero data retention as a $1,000/mo add-on. If you handle patient data on a budget, this single row decides your shortlist.
- Resilience: published SLAs and failover behavior are thin across all three vendors. This is the gap our follow-up lab test will measure [UNVERIFIED].
Adoption and lock-in (points 14 to 15)
- Learning curve: Synthflow is buildable by a non-technical operator in an afternoon. Retell sits in the middle with visual flows plus an API. Vapi assumes you have engineers, and rewards you for it.
- Reversibility: weak everywhere. None of the three offers a portable export format for agent logic. Your prompts and flows are rebuildable, your call data is exportable, but migration means reconstruction. Plan for it from day one.
04Verdict by team profile
Decision rule: your engineering capacity picks the platform. No engineers, pick Synthflow. Some engineers and a deadline, pick Retell. A real dev team that wants control of every component and every cent, pick Vapi.
| Team profile | Recommendation |
|---|---|
| Agency or non-technical operator deploying agents for clients, white-label needed | Synthflow |
| SMB or startup that needs production quality fast, with predictable bundled pricing | Retell AI |
| Engineering team optimizing cost and model choice at scale, comfortable owning the stack | Vapi |
| Healthcare on an entry-level budget | None qualify. HIPAA is Enterprise-gated or a paid add-on across all three. |
How we audit, who pays us, and what would get a platform excluded from this page: methodology and disclosure.