Vapi vs Retell AI

Reachkit makes cold email software, not voice agents, so it competes with neither tool here. Every price on this page was checked in October 2026 against Vapi’s and Retell AI’s own pages and docs.

My verdict: both build AI agents that answer and place phone calls, and the split is control against packaging. Vapi is API-first: you pick the transcriber, model, voice and carrier, and it bills those providers at their listed rates plus $0.05 a minute for hosting. Retell AI charges $0.055 a minute for infrastructure that, by its account, includes speech-to-text, sets its own model and voice prices, and includes 20 concurrent calls. A real minute costs about $0.08 to $0.13 on Vapi’s calculator before carrier fees, and about $0.11 to $0.20 on Retell, from its calculator and my math on its rate card.

Pick Vapi if:

• Someone on your team writes code and wants to swap any provider, or bring their own API keys and pay providers directly.
• You want a model from Groq, xAI or DeepSeek; Retell’s rate card lists OpenAI, Anthropic and Google, plus your own LLM.
• You want every call’s cost broken down by component, and recordings kept in your own cloud storage.

Pick Retell AI if:

• The people building the agent think in flows, not code, and want an AI helper that proposes edits.
• You need many calls at once without buying a package.
• You want phone numbers and US calling from the same vendor: $2 a month per number, $0.015 a minute.

Checked October 2026, from each vendor’s public pages and docs. Prices in USD.

Vapi hands you the parts and Retell AI hands you a builder, though each moved toward the other during 2026.

On Vapi you create an assistant and choose a transcriber, a language model and a voice, either one by one or from a Model Preset such as Balanced or Cost Saver (added in July 2026). You write the prompt, attach tools and connect a phone number or a web widget. Calls that need several roles run as Squads, which pass the caller between assistants and gained a visual builder in Squads v2. Composer, an AI helper in the dashboard, builds, adjusts and debugs agents from text prompts. Vapi’s older Workflows product stopped running on August 19, 2026.

Retell AI starts from a single prompt, several prompts, or a drag-and-drop conversation flow whose nodes handle functions, transfers, keypad input and SMS. Flex mode, added in December 2025, lets a caller jump between nodes. Conductor, launched in July 2026, builds agents and proposes edits for you to approve; each user gets 30 free credits a day, then $0.20 a credit if an admin turns on paid use.

The tradeoff is who owns the agent. Vapi’s choices pay off when an engineer does; Retell’s flows are easier to hand to someone who doesn’t write code.

Retell AI includes more calls at once without a package, and Vapi asks for more phone setup before an agent can dial out.

Concurrency, the number of calls running at the same moment, decides how fast a list gets worked. New Vapi accounts get 4 without a package, 10 on Core ($29 a month) and 30 on Pro, with extra lines at $10 a month each; accounts on Vapi’s legacy pricing keep 10 per organization. Vapi’s docs suggest feeding lead lists in chunks of 50 to 100 numbers. Retell includes 20 on pay as you go, then $8 a month per extra concurrent call. With burst mode turned on, calls above the limit can still connect, up to three times the limit or the limit plus 300, whichever is lower, with a $0.10 a minute surcharge on the whole call.

Numbers differ too. Vapi’s free number has a US area code, takes inbound calls only and can’t be ported out, so dialing out means importing a number from Twilio, Telnyx, Vonage or another carrier. Retell sells numbers at $2 a month with US calls at $0.015 a minute, charges nothing on your own SIP trunk, and adds branded caller ID at $0.10 per outbound call.

For campaigns, Vapi’s dashboard (since August 31, 2026) takes a CSV of up to 10,000 contacts, schedules up to seven days ahead and detects voicemail; a campaign can’t be edited once created, only duplicated. Retell’s batch calls cost $0.005 a dial and run no faster than your concurrency and calls-per-second limits allow.

Both platforms let you test an agent before it takes real calls; they differ on support and paperwork.

Vapi has simulated test calls, evals, monitoring, and, since August 2026, versioning with drafts and rollback. Each call log breaks its cost down by component. Retell has simulation and batch tests, live listen-in with takeover, A/B tests, and versioning with staging and production since May 2026. Retell’s docs say test calls bill at production rates with no free sandbox, and its AI quality reviews are free for the first 100 minutes, then $0.10 a minute.

Without a paid tier, Vapi support is Discord plus email with no SLA; Core adds a two-business-day SLA, and Pro a one-day SLA with Slack. Retell’s pay as you go has community and email support, and 24/7 support is on Enterprise.

On compliance, Vapi’s docs say its SOC 2 Type II report covers all customers. HIPAA is a $2,000 a month add-on per organization with a BAA, zero data retention starts at Core, and SSO and an EU region are Premier only. Retell’s homepage says it is compliant with HIPAA, SOC 2 Type II, GDPR and ISO 27001, and its comparison page with Vapi (data as of May 2026) says those cover every plan. Its pricing page lists custom MSA and DPA terms and SSO under Enterprise. If you need a signed BAA, ask Retell which plan includes one.

Vapi supports the wider range of model providers, and both are open to developers through an API and an MCP server.

Vapi works with models from OpenAI, Anthropic (including through Amazon Bedrock), Google Gemini, Groq, xAI, DeepSeek and others, plus custom LLM endpoints. Voices come from ElevenLabs, Cartesia, Deepgram, PlayHT and Vapi’s own, and with your own provider keys those providers bill you directly. Call recordings can go to your own S3, Google Cloud, Cloudflare R2, Azure or Supabase storage.

Retell offers OpenAI, Anthropic and Google models in a standard tier or a fast tier at roughly twice the price, your own LLM, and OpenAI’s speech-to-speech Realtime models. Voices come from Retell, ElevenLabs, Cartesia, OpenAI, Minimax, Fish and Inworld.

Both publish a REST API, SDKs and webhooks, and both host an MCP server, a standard way for AI assistants like Claude and ChatGPT to use other apps. Vapi’s lets an assistant manage agents, numbers and calls; Retell’s reaches its whole API through three tools. Agents on both can call outside MCP tools during a call. Retell sets up two-way contact sync with Salesforce, HubSpot, Dynamics 365, GoHighLevel and Zoho, and added a built-in CRM in June 2026; Vapi lists HubSpot, Salesforce, GoHighLevel, Zapier and Make among its integrations.

The core difference in this pair is who sets the component prices: Vapi passes providers through at their own rates, and Retell AI publishes a rate card of its own.

A Vapi minute is $0.05 of hosting plus the transcriber, the model and the voice, each “billed at the provider’s listed price with no markup”, in the words of Vapi’s pricing FAQ. Transport over Vapi’s own numbers is free; Twilio charges $0.014 a minute for outbound. The model is billed by tokens, and Vapi’s docs warn that long calls cost more because the conversation history is sent again with each request.

A Retell minute is $0.055 of voice infrastructure plus a voice and a model at Retell’s listed rates. Standard voices are $0.015 a minute, ElevenLabs runs $0.04 to $0.10, and GPT 5.1 is $0.04. Vapi’s calculator shows ElevenLabs at about $0.015 to $0.024 a minute, though it doesn’t name the ElevenLabs model. Retell prices the model per minute, but prompts past 4,000 tokens, counting the instructions, tools, knowledge base and conversation so far, stretch the billed time in proportion.

So a long, chatty call costs more on both, by different routes. Retell’s card is easier to forecast; Vapi’s pass-through comes out lower wherever Retell’s listed rate sits above the provider’s own.

Both bill by the minute from prepaid credits, so the fair comparison is a month of US calls at public prices (October 2026).

Assumptions: Vapi uses its calculator’s default ranges for Deepgram, OpenAI and ElevenLabs, Twilio outbound at $0.014 a minute, and Core ($29) for 10 lines. Retell’s first row is its calculator default of $0.11 a minute; the others run from GPT 5.1 with a Retell voice ($0.125 a minute) to GPT 5.4 with ElevenLabs Flash and a knowledge base ($0.195), on one $2 Retell number. Vapi’s calculator doesn’t name its model, so treat the rows as a guide, not a quote. Retell’s batch campaigns add $0.005 a dial.

Concurrency changes the picture at volume. Twenty calls at once cost nothing extra on Retell; on Vapi that’s Core plus ten lines, $129 a month, or Pro at its $999 minimum. Below 10 calls at once, your model and voice choices move the bill more than the platform does.

Retell AI has many more public reviews than Vapi, and Vapi’s small samples disagree.

Vapi (October 2026): 2.4 from 15 reviews on Trustpilot, on a profile Vapi hasn’t claimed, against 4.9 from 25 on Product Hunt, 18 of them from founders. G2 holds only a handful. Builders praise the choice of providers, and one Product Hunt reviewer wrote that “the API is easy to integrate”. Trustpilot complaints from December 2025 to March 2026 cover pricing one reviewer called “miss leading”, pauses between turns and agents drifting from the prompt. Vapi shipped versioning and simulated calls in August 2026, after those reviews.

Retell AI (October 2026): 4.9 from 810 reviews on Trustpilot, many of them invited, and about 4.8 from roughly 2,600 on G2 per search results. Reviewers praise quick setup (“So easy to use”, June 2026) and call quality. Trustpilot’s few complaints include support: a March 2026 reviewer found it “rather slow”, and a July 2026 review described batch calls not going out with “no one here to help”. Older summaries of G2 reviews flag cost for small teams and a learning curve.

A Reddit thread from about a year ago puts the split the way most top-ranking comparisons do: one poster found Retell AI more “plug and play” for real estate teams, and said Vapi “looks promising if you want finer control over the voice itself”.

This is our product, and it doesn’t make calls, so it replaces neither tool. It fits one reader here: someone who hoped a voice agent would work a cold list. Given the US consent rules in the FAQ below, I’d start those conversations by email and point the voice agent at people who answered. Reachkit finds verified leads among 500M+ profiles at 1 credit each, drafts the sequence from one sentence, registers domains ($15 a year) and sets up and warms inboxes ($5 a month), and labels replies such as Meeting Requested. It doesn’t answer or book meetings for you. Growth is $39 a month per workspace; on Scale, from $99, a webhook fires when a lead replies.

Moving between Vapi and Retell AI means rebuilding the agent, but carrier numbers, hosted tools and your CRM come with you.

• Phone numbers. Numbers you hold at Twilio, Telnyx or Vonage connect to either platform, and Retell charges nothing on your own SIP trunk. Vapi’s free numbers can’t be ported out, so callers who know one will need a new number.
• The agent. Copy the prompt, then choose the model and voice again, since the two price lists don’t map one to one. Plan to rebuild Squads or conversation flows by hand.
• Tools. Webhook endpoints and MCP tools you host keep working; point the new agent at them and reconnect the CRM.
• Credits. Both treat usage as non-refundable. Retell’s credits never expire; Vapi’s terms and docs don’t say.
• Legacy Vapi accounts keep their old terms, including 10 concurrent lines where a new account starts at 4, so weigh that before closing one.

Run the same script on both for a week before moving live numbers, and budget for the test minutes.

Both are serious platforms, and the choice comes down to who on your team will own the agent. If a developer wants to pick each provider, read the cost of every call component and bring their own keys, I’d start on Vapi and buy Core once 4 calls at once gets tight. If the people building it think in flows rather than code, or you need 20 calls running in parallel from the start, Retell AI asks less of them. Price both on your real call length, not the headline rate. And if the real goal is reaching cold prospects, start with email, which is what Reachkit does.

More detail: the Vapi profile, the Retell AI profile, Vapi alternatives and Bland alternatives.