Vapi Review (2026)

Who Vapi is for

Developers who want control over every component of the voice stack. Vapi is the choice when you have opinions about which speech model, which voice and which telephony carrier to use, or when you already hold contracts and API keys with those vendors and want to reuse them.

What we verified

We checked vapi.ai/pricing on September 7, 2026. The Build plan is pay-as-you-go with no monthly platform fee and '60+ minutes included'. Vapi charges a $0.05 per-minute platform fee; speech-to-text, the language model and text-to-speech are billed at cost, or $0 if you bring your own API keys; telephony is charged by your provider. Build includes 10 concurrent lines, with additional lines at $10 per line per month, and SMS or chat at $0.005 per message. A Scale plan pairs a fixed platform fee with committed volume and volume-based per-minute pricing. A HIPAA add-on costs $2,000 a month and zero-data-retention $1,000 a month.

Strengths

• Bring-your-own models and voices, cutting pass-through cost to $0 with your own keys.
• Transparent $0.05 per-minute platform fee.
• Strong developer tooling and SDKs (not independently verified).
• Recording, transcripts, outbound calling and multi-language (not independently verified).

Limitations

• Not a turnkey receptionist.
• Model and telephony costs stack on top of the platform fee.
• HIPAA compliance costs $2,000 a month extra.
• No small-business workflow out of the box and no human backup.

How we evaluated it

This review is based on Vapi's pricing page, checked September 7, 2026, and the positioning recorded in our provider database, judged against the expectations for a developer platform rather than a packaged receptionist. Vapi has not completed the 10-call challenge. Lab testing will follow the same approach we use for Retell: a documented reference build, so that platform behavior rather than builder skill is what gets scored.

What the product is designed to do

Vapi positions itself around choice. You select the speech and language models, the voice and the telephony provider, then orchestrate the agent through APIs and SDKs. Recording, transcripts, outbound calling and multi-language support are listed in its materials; scheduling and CRM are marked partial because they depend on tools you implement. There is no human backup, no vertical template and no small-business workflow out of the box. The design target is an engineering team with opinions about every layer of the voice stack and the freedom to act on them.

Setup and day-to-day operation

Expect a build comparable to Retell's, with an additional layer of decisions about which components to use. Each choice of model, voice or carrier affects quality, latency and cost, and those choices have to be revisited as vendors change. The team must implement booking and lookup tools, handle transfers and build the reporting the business needs. Ongoing operation includes monitoring, prompt maintenance and watching how the separately billed components add up: the $0.05 platform fee is the floor, and with third-party models billed at cost the effective rate lands well above it unless you bring your own keys. Managing that stack is the main operational discipline a Vapi deployment demands.

Where it fits in our ranking

Vapi is ninth. It ranks below Retell not because it is less capable, but because its emphasis on maximum configurability pushes more decisions onto the buyer, and because Retell publishes an all-in per-minute range while Vapi's total depends on components you choose. For most businesses reading a receptionist ranking that is a cost rather than a benefit. It ranks above Slang AI only because Slang serves a single vertical outside our core audience. Teams choosing between the two platforms should read Retell vs Vapi; those unsure whether they need a platform at all should start with Synthflow vs Retell.

Which businesses should shortlist it

Product teams and agencies with the engineering depth to exploit bring-your-own components. A software vendor building a scheduling product for plumbing companies or electrical contractors fits; the contractors themselves do not, and should shortlist a packaged receptionist instead.

What would change our view

• Lab results from a reference build. Strong performance would confirm that the flexibility does not come at the expense of call quality.
• A measured all-in per-minute cost for the reference build with common third-party components, to place beside Retell's published range.
• Evidence on the time required to reach a working receptionist, which decides whether maximum customization is a benefit or a burden for a given team.
• A confirmed integration and tooling catalog.