Vapi review
The most flexible, developer-first voice-agent platform — power at the price of assembly.
The most flexible voice-agent platform on the market, but priced and packaged for developers — not for a non-technical owner who wants a receptionist live by lunch.
Best for: Developers and technical teams who want maximum control over the voice stack and are comfortable assembling and tuning STT, LLM, TTS and telephony themselves. · From: Build (pay-as-you-go): $0.05/min platform fee + pass-through STT/LLM/TTS/telephony (model costs $0 if you bring your own API keys); ~$0.15–$0.30/min all-in typical. 10 concurrent call lines included, then $10/line/mo. SMS/chat $0.005/msg. $10 free signup credit, no ongoing free tier. Add-ons: HIPAA $2,000/mo, zero-data-retention $1,000/mo. Scale plan is custom/volume-based (contact sales).
Pros
- + Bring-your-own everything: pick any STT, LLM, TTS and telephony provider, so you're never locked into one vendor's voice or model
- + Lowest headline platform fee in the category at $0.05/min, and $0 model markup if you supply your own provider API keys
- + Genuinely capable at scale — API-first, SDKs, workflow builder, real-time monitoring, inbound and outbound, sub-500ms advertised latency
- + Compliance options (SOC 2, HIPAA, PCI) exist for regulated use cases
Cons
- − Headline $0.05/min is misleading — real all-in cost is typically $0.15–$0.30+/min once STT, LLM, TTS and telephony stack on top
- − Steep for non-developers: you wire together and debug four separate providers plus phone numbers, not a turnkey setup
- − Base plan support is community Discord and email only; dedicated help requires the enterprise Scale plan
- − Only a $10 signup credit (~30–60 test minutes) and no ongoing free tier; compliance add-ons are expensive ($2,000/mo HIPAA, $1,000/mo zero-data-retention)
Setup
Vapi is API-first and assumes you are a developer. You create an assistant, then wire up four moving parts yourself — speech-to-text, an LLM, text-to-speech, and a telephony/number provider — and optionally supply your own provider API keys to avoid markup. There is a dashboard and a no-code workflow builder, but the real power, and the real debugging, lives in the SDKs and API. A developer can have a working agent live in an afternoon; a non-technical owner will not get there alone.
The AI / voice quality
Because Vapi is provider-agnostic, voice quality is whatever you choose — ElevenLabs, Cartesia, Deepgram, PlayHT and 25+ integrations. Pair a fast STT with a low-latency TTS and Vapi advertises sub-500ms voice-to-voice latency; independent tests more commonly report 550–800ms depending on model and region. It handles inbound and outbound calls, interruptions, function calling and real-time monitoring. The ceiling is high — you are responsible for the tuning it takes to reach it.
Pricing
The $0.05/min platform fee is the lowest headline in the category, and model cost is $0 if you bring your own STT/LLM/TTS API keys — otherwise those layers pass through at cost. Realistically the all-in figure lands around $0.15–$0.30/min once every provider and telephony charge stacks up, so $0.05 is not what you actually pay. Ten concurrent lines are included, then $10/line/mo; SMS/chat is $0.005/msg. New accounts get $10 in credit (roughly 30–60 test minutes) but there is no ongoing free tier. Compliance is a hard cost: HIPAA is $2,000/mo and zero-data-retention $1,000/mo. The Scale plan is a custom annual contract with a fixed platform fee plus committed volume.
Free / cheaper alternatives
- Retell AI — more turnkey, simpler bundled per-minute pricing, and far easier for a non-developer to stand up.
- Bland AI — a fully managed voice stack where you do not assemble providers yourself; strong for outbound.
- Vocode (open source) — self-host the entire pipeline for effectively free if you have the engineering time to run it.
Verdict
Vapi is the right call when control is the point: you want to swap models freely, drive everything from code, and squeeze cost with your own API keys. It is the wrong call if you want a receptionist working by lunch without touching four vendors and a phone-number config. The $0.05/min headline is real but incomplete — budget $0.15–$0.30/min and community-only support on the base plan.