Vapi is a developer platform for building voice AI agents — the bots that pick up the phone, book appointments and qualify leads in real conversations. Its pitch is flexibility: you choose the speech-to-text, the language model and the voice, and Vapi orchestrates the whole pipeline behind one API. That model-agnostic control makes it a favorite with engineers, but the same freedom means costs stack up from several providers at once, starting around a $0.05/min platform fee.
Rating: 4.3/5
Verdict: Vapi is the most flexible way to build production voice agents if you can code — unmatched model choice and low-latency orchestration — but it's a developer tool, and the multi-provider billing means you have to actively manage cost.
Best for: Engineering teams building high-volume phone and web voice agents who want full control over models and voices.
What is Vapi?
Vapi is voice-agent infrastructure. Rather than shipping a finished chatbot, it gives developers the plumbing to assemble one: a single API (with SDKs for web, iOS and JavaScript) that connects speech recognition, an LLM brain, a text-to-speech voice and telephony into a coherent, low-latency call. You bring the pieces; Vapi makes them talk to each other in real time.
That places it differently from the tools in our best AI voice generators roundup. Most of those — like ElevenLabs — focus on generating the voice. Vapi focuses on running the conversation: turn-taking, interruptions, tool calls and phone connectivity, with the voice as just one swappable component.
How the pipeline works
A Vapi agent chains four layers. Speech-to-text (from providers like Deepgram or AssemblyAI) transcribes the caller; a language model (OpenAI, Anthropic or others) decides what to say and which tools to call; text-to-speech (ElevenLabs, PlayHT and more) speaks the reply; and telephony (typically Twilio) carries the call. Vapi manages the timing so the exchange feels natural rather than walkie-talkie.
The upside of this design is total control — swap any layer to trade cost against quality, wire in your own functions and databases, and tune latency aggressively. The downside is that each layer bills separately, so your real per-minute cost is the sum of all four plus Vapi's fee. Understanding that math up front is essential.
Features that stand out
Beyond raw orchestration, Vapi does the unglamorous production work well. Agents live in a workspace with access controls, memory, integrations and evaluations, so a team can train, deploy and update bots from one place. Built-in evals matter more than they sound — testing voice agents is hard, and being able to measure quality across calls is a real advantage.
It's also tuned for scale. Because it's model-agnostic and leans on external providers, a well-configured Vapi setup can support large-scale inbound and outbound calling without hitting a ceiling. For contact-center-style workloads, that headroom is a major reason teams pick it.
Vapi pricing in 2026
Vapi's pricing is usage-based and layered, which is the single most important thing to understand before you build on it.
| Component | Typical cost | Notes |
|---|---|---|
| Vapi platform fee | ~$0.05/min | The orchestration layer only. |
| Speech-to-text | Provider rate | Deepgram, AssemblyAI, etc. — billed separately. |
| LLM | Per-token | OpenAI, Anthropic and others; depends on model + length. |
| Text-to-speech | Provider rate | ElevenLabs, PlayHT, etc. |
| Telephony | ~$0.01/min per leg | Twilio or similar. |
| Realistic all-in | ~$0.07–$0.25/min | Higher with premium voices and heavier models. |
The advertised $0.05/min is only Vapi's slice. In practice, most production agents land between $0.07 and $0.25 per minute once STT, LLM, TTS and telephony are added, and premium voices with a top-tier model can push it higher. New accounts get about $10 in free credit to prototype, but there's no ongoing free tier. The upside: for high-volume bots that can use good-enough voices, this BYO model can undercut all-in platforms.
Pros and cons
After building and running test agents on it, here's the ledger.
Pros
- Fully model-agnostic — pick any STT, LLM and voice
- Low-latency orchestration that feels natural on calls
- Built-in workspace, memory, integrations and evals
- Scales to high-volume inbound/outbound calling
- Can undercut all-in platforms on cost when tuned
Cons
- Developer-only — expects code, not a no-code UI
- Multi-provider billing makes true cost hard to predict
- No ongoing free tier beyond ~$10 trial credit
- You own the tuning of latency and quality
- Premium voices erase much of the cost advantage
Who Vapi is for
Pick Vapi if you're an engineering team building voice agents at scale and you want control over every layer of the stack. For custom phone bots, support lines and outbound campaigns where you'll tune models and manage cost deliberately, it's one of the best foundations available.
Skip it if you're non-technical or want something that works out of the box — an all-in-one voice product will get you there faster. And if premium voice realism is the priority, weigh it against ElevenLabs Agents in Vapi vs ElevenLabs, read our ElevenLabs review, and browse the wider field in the best ElevenLabs alternatives.
Rated 4.3/5, Vapi is the developer's voice platform — enormously capable in the right hands, and demanding of them in return.
Frequently Asked Questions
What is Vapi?
Vapi is a developer platform for building voice AI agents. It ties together speech-to-text, a large language model, text-to-speech and telephony into one API, letting you build phone and web voice bots that handle real conversations.
How much does Vapi cost?
Vapi charges a platform fee of about $0.05 per minute for orchestration, then passes through the cost of the STT, LLM and TTS providers you choose, plus telephony. A typical deployment runs roughly $0.07-$0.25 per minute all-in. New accounts get $10 in free trial credit.
Is Vapi good for non-developers?
Not really. Vapi is built for engineers — its power comes from model choice and API control, which means it expects code. Non-technical teams are usually better served by an all-in-one voice agent product that trades flexibility for a no-code interface.
What can you build with Vapi?
Inbound and outbound phone agents, web voice assistants, appointment booking, lead qualification, customer support lines and voice-driven workflows. It's well suited to high-volume calling once the pipeline is tuned.
Is Vapi cheaper than ElevenLabs Agents?
It can be. Because you bring your own STT, LLM and TTS, Vapi can be 30-40% cheaper than ElevenLabs Agents for high-volume bots that don't need premium voices. The trade-off is more setup and cost management on your side.
Does Vapi have a free tier?
There's no ongoing free tier, but new accounts receive about $10 in credit to test the platform. After that you pay per minute plus the underlying provider costs.