Vapi Overview
This Vapi review covers the voice AI platform as it stands in 2026. Vapi is the developer’s voice agent platform: instead of a polished no-code builder, it gives you orchestration infrastructure for phone agents where you bring your own language model, speech-to-text, text-to-speech, and telephony — and Vapi coordinates them with sub-500ms latency. With a reported 350,000+ developers and 150 million+ calls processed, it is one of the most-used foundations for production voice AI. But the headline price hides the real bill, so this review spends extra time on what Vapi actually costs.
What is Vapi?
Vapi is an API-first platform for building, testing, and deploying conversational voice agents over the phone and the web. You define an assistant via JSON configuration or the dashboard: pick a model (GPT-4o, Claude, Gemini, Mistral, or your own endpoint), a voice provider (ElevenLabs, Deepgram, Cartesia, PlayHT, Azure), a transcriber, and a phone provider (Twilio, Vonage, or Vapi’s own numbers). Vapi handles the hard real-time parts — audio streaming, interruption detection, turn-taking, endpointing — and exposes webhooks and a REST API for everything else: call routing, data extraction, CRM updates. Calls can be inbound or outbound, and SIP trunking scales to many concurrent conversations.
Key features
- Provider-agnostic orchestration: mix and match any STT, LLM, TTS, and telephony provider independently; bring your own API keys.
- Sub-500ms latency target: documented engineering investment in turn-taking, endpointing, and background-noise filtering.
- Squads: multi-agent orchestration where specialized agents hand off within one call with silent transfers.
- Composer: visual workflow builder for teams that do not want to hand-code every flow.
- Automated testing: simulate calls in chat or voice mode with exact-match, regex, or AI judges; CI/CD friendly.
- Structured outputs: extract validated, typed data from calls into your systems.
- Developer tooling: CLI, MCP server, monitoring dashboard, and Python/JS SDKs.
- Vapi Voices: native TTS option launched to cut cost and latency versus third-party voices.
- Compliance options: SOC 2, HIPAA, and zero-data-retention available as add-ons.
What Vapi actually costs
This is the section that matters most. Vapi’s published price is $0.05 per minute — but that is only the orchestration fee. STT, the LLM, TTS, and telephony are pass-through costs billed by their providers (or free if you attach your own keys, in which case you pay those vendors directly). Independent reviewers consistently put the realistic all-in figure at $0.13-$0.31 per minute depending on provider choices: for example, Deepgram Nova-3 STT (~$0.007/min) + GPT-4o (~$0.005/min) + ElevenLabs TTS (~$0.02/min) + the $0.05 floor lands around $0.08-$0.15/min before telephony. New accounts get $10 in free credits, roughly 150-200 test minutes. The Build plan is pay-as-you-go; Scale is an annual contract with SOC 2, SSO, and bundled provider billing. The HIPAA add-on is steep — around $2,000/month, plus a $1,000/month zero-data-retention option — which makes regulated use cases dramatically more expensive than the headline suggests. Model the full stack before committing; the demo math is not the production math.
Who is it for?
Vapi is for engineering teams that want maximum control over their voice stack and are comfortable owning provider selection, prompt engineering, and cost optimization. If you need to swap models independently — say, Claude for reasoning but Cartesia for voice — no competitor makes that easier. It is a poor fit for non-technical teams: there is no true no-code builder, branded caller ID is absent (unlike Retell), and you will write real integration code. Sales teams wanting outbound campaigns running by Friday should look at purpose-built outbound platforms instead.
Strengths and honest weaknesses
The flexibility is unmatched: any model, any voice, any transcriber, your own keys, your own endpoints. The latency engineering is serious — custom endpointing models, background voice filtering, and emotion-aware context are differentiators, not marketing. The weaknesses are cost transparency and operational burden. The unbundled model means your bill arrives from four vendors and forecasting is genuinely hard at volume. Third-party reviewers note a 2.6 Trustpilot rating with complaints concentrated on pricing surprises. Squads, the flagship multi-agent feature, is described as buggy in practitioner reports. And HIPAA pricing puts regulated startups in a tough spot.
Getting started
Claim the $10 in free credits and build a minimal assistant in the dashboard before writing code — pick Deepgram for STT and Vapi Voices for TTS to keep the first test cheap. Use the automated testing framework from day one; voice agents fail in ways chatbots do not, and simulated calls catch prompt issues fast. Track all-in cost per minute from the first real calls, not the $0.05 floor, and decide your provider stack on measured numbers.
Alternatives to consider
Vapi’s main rivals frame the trade-offs clearly. Retell AI offers lower out-of-the-box latency (~600ms), a visual agent builder, and branded caller ID that keeps calls from being flagged as spam — at $0.07-$0.23/min platform fees but less provider flexibility. Bland AI owns its full model stack for tighter data control and simpler single-vendor billing. LiveKit and Pipecat serve teams that want open-source orchestration with maximum control and are willing to do more engineering. Synthflow targets non-technical teams with no-code builders and 200+ integrations. Choose Vapi when provider flexibility and cost optimization are the priorities; choose Retell when you want the fastest path to a production-quality agent with less engineering.
Verdict
Vapi is the best foundation for developers who want to own every layer of their voice agent and optimize cost and latency themselves. It is the worst choice if you want predictable single-invoice billing or a no-code experience. Go in with eyes open about the real per-minute cost, and it is a genuinely powerful platform.
Key Features
- Provider-agnostic orchestration: any LLM, STT, TTS, and telephony, BYO keys
- Sub-500ms latency with custom endpointing and background-noise filtering
- Squads multi-agent handoffs with silent transfers in one call
- Composer visual workflow builder plus full JSON/API configuration
- Automated call testing with AI judges and CI/CD integration
- Structured outputs: validated typed data extraction from calls
- Vapi Voices native TTS for lower cost and latency
- CLI, MCP server, monitoring dashboard, Python/JS SDKs
Vapi Pricing
| Plan | Price |
|---|---|
| Free trial | $10 in credits (~150-200 min) |
| Build | $0.05/min + provider costs |
| Scale | annual contract, custom |
| HIPAA add-on | ~$2,000/mo |
| Zero data retention | ~$1,000/mo |
Pricing checked on November 2, 2026 — always confirm on the official site.
Vapi Pros & Cons
✓ Pros
- Unmatched provider flexibility — swap any model, voice, or transcriber independently
- Serious latency engineering: custom endpointing, filler injection, noise filtering
- Automated testing framework catches voice-agent failures early
- 350k+ developers and 150M+ calls of production proof
- Vapi Voices offers a path to the lowest per-minute cost at scale
✕ Cons
- The $0.05/min headline is misleading — real all-in cost runs $0.13-$0.31/min once STT, LLM, TTS, and telephony are added.
- Requires serious engineering effort; there is no simple no-code path for non-technical teams.
- Squads multi-agent handoffs are powerful but reportedly buggy and hard to implement reliably.
- HIPAA compliance demands an expensive $2,000/month add-on, punishing for small healthcare teams.
- Thin support below enterprise tiers, and third-party reviews flag pricing transparency complaints.
Vapi FAQs
What is Vapi used for?
How much does Vapi cost per minute?
Is Vapi better than Retell AI?
Does Vapi support HIPAA compliance?
Can non-technical users build agents on Vapi?
Best Vapi Alternatives
Foley
★★★☆☆Foley review (2026): a voice to-do list that files tasks from speech and asks before guessing — free desktop app, $8/mo Pro. Features and honest verdict.
Cekura Bench
★★★★☆Cekura Bench reviewed for 2026 — the open speech-to-speech benchmark testing voice AI models on live phone calls. Features, pricing, verdict.
Altered
★★★★☆Altered reviewed for 2026 — AI voice cloning, morphing, and TTS from $30/mo. Voice quality, real pros and cons, and who should actually pay for it.
Synthesys
★★★☆☆Synthesys reviewed for 2026 — 1,000+ AI voices and avatar videos in one studio. Worth $20/mo? Features, pricing, and honest verdict.