comparison · infra layer
Vapi vs Retell in 2026: Pricing, Features, Latency, Real Tradeoffs
By builders, for builders.
Vapi and Retell are the two most-used infrastructure layers for production AI voice agents in 2026. Both ship a similar core: a hosted runtime that wires an LLM, a TTS engine, an STT engine, and a telephony provider into a single low-latency duplex stream. Both publish a usage-based price in the $0.05 to $0.31 per minute range depending on configuration. Both have public APIs, SDKs in TypeScript and Python, and a hosted dashboard. The differences sit at the edges: how opinionated the flow builder is, how the dashboard surfaces multi-tenant ownership, what passthrough vendor costs look like, and which provider catalog you can swap in. This page is the honest side-by-side. The short version: pick Vapi if you want maximum dev flexibility and serverless function-call patterns. Pick Retell if you want a slightly more opinionated builder with built-in voice models. If you are running an agency with 5 or more voice clients, you will outgrow both at the multi-tenant layer.
Side-by-side comparison
| Capability | Vapi | Retell |
|---|---|---|
| Pricing model | Per-minute platform fee + passthrough | Per-minute bundled or BYO-keys discount |
| Included minutes | None, pure usage | None, pure usage |
| Per-minute all-in cost | $0.13 to $0.20 | $0.07 to $0.31 |
| Built-in CRM | No | No |
| White-label portal | No turnkey | No turnkey |
| Multi-tenant client billing | Build it yourself | Build it yourself |
| Voice models in catalog | BYO + integrations | Native + BYO |
| Latency p50 | ~700 to 900 ms | ~600 to 850 ms |
| Latency p95 | ~1.4 to 1.8 s | ~1.2 to 1.6 s |
| Integrations style | Function calls (serverless) | Webhooks + custom functions |
| Support tier | Self-serve + paid plan support | Self-serve + paid plan support |
| Agency-friendly out of box | No, infra layer | No, infra layer |
| SOC 2 / HIPAA available | Available on enterprise plans | Available on enterprise plans |
| Funding stage | Seed / Series A range | Series A range |
Best for: Vapi
Pick Vapi if your team is dev-first and you want maximum control over the function call surface. Vapi's function-call model gives clean serverless patterns where the LLM hands off to your endpoint, your endpoint does work, returns a structured response, and the conversation resumes. If you are building a single in-house voice product for one company and you have engineers comfortable with that pattern, Vapi is a strong pick. See the Vapi documentation for current pricing and provider catalog.
Best for: Retell
Pick Retell if you want a slightly more opinionated builder with a stronger out-of-the-box voice catalog and a flow-style UI for prompt composition. Retell tends to be the faster path to a working agent for non-dev operators because the visual flow surfaces decisions the dev would otherwise hand-code. Retell's BYO-keys discount also makes it cost-attractive at scale if your team already manages OpenAI and ElevenLabs accounts. See the Retell documentation for current pricing.
Where Hermes fits if you outgrew both
If you are running 1 client, either Vapi or Retell is fine. If you are running 5 or more clients as an agency, the multi-tenant layer is where Vapi and Retell both leave you to build your own. That means Stripe Connect for per-client billing, your own CRM (probably GoHighLevel), Zapier or n8n for glue, custom dashboards for your clients, and a white-label portal you maintain. Hermes is the operating layer that handles all of that on top of the same voice providers. One workspace per client, native CRM, transparent voice overage at $0.24 per minute flat, and white-label demo pages bound to your own CNAME.
| Capability | Vapi or Retell alone | Hermes |
|---|---|---|
| Agency-tier plan | N/A, you build the agency layer | $699/mo · 20 workspaces · 2,000 min pooled |
| Per-minute overage | $0.13 to $0.31 + complexity | $0.24 flat |
| Native CRM | No, bring GHL or build | Native contacts, pipeline, sequences |
| White-label demo pages | Build them yourself | CNAME-bound, included on Business+ |
| Multi-tenant client billing | Stripe Connect yourself | Per-workspace P&L native |
Related: Hermes vs Synthflow, Hermes vs the Vapi + GHL stack, Hermes vs Voicerr.
FAQ
Is Vapi cheaper than Retell?
On a per-minute basis they sit close. Vapi runs roughly $0.05 to $0.07 per minute for the platform fee plus passthrough provider costs (LLM, TTS, STT, telephony) which usually total $0.13 to $0.20 per minute end-to-end. Retell runs $0.07 to $0.31 per minute depending on the voice model tier and whether you bring your own LLM keys. For most production agencies the all-in cost lands within 10 percent of each other. The actual cost driver is provider model choice, not the underlying platform.
Which has lower voice latency, Vapi or Retell?
Both publish p50 latency in the 700 to 900 ms range under default configurations. Retell tends to do better on first-token latency when configured with their built-in voice models. Vapi tends to do better when you supply your own faster TTS like ElevenLabs Flash. The honest answer is that both are within human-conversational tolerance and your audible latency will be dominated by your network path and provider choice, not the platform.
Can I white-label Vapi or Retell for my agency clients?
Vapi has a hosted dashboard your clients can log into, but the brand and chrome are Vapi. Retell similarly offers an API and a dashboard, with limited theming. Neither ships a turnkey white-label portal where your client logs in and sees your agency logo, your domain, and your billing. To get that you either build the portal yourself or use an aggregator platform like Hermes that runs on top of voice infra and handles the white-label layer.
Which is better for building an AI voice agency, Vapi or Retell?
Neither is built for the agency business model. Both are infrastructure layers: you bring code, glue, billing, CRM, and white-label yourself. Vapi tends to attract devs who want maximum control and are comfortable with serverless function call patterns. Retell tends to attract teams who want a slightly more opinionated platform with built-in flow building. For a single agency client they are both fine. For 5+ clients with multi-tenant billing and white-label, agencies usually layer something on top.
Does Hermes replace Vapi or Retell, or work alongside them?
Hermes is the agency operating layer that sits on top of voice infrastructure. We use the same providers underneath (Retell, Vapi-style stack components) but expose them through one workspace per client, with native CRM, native multi-tenant billing, and white-label demo pages on your domain. Agencies on Hermes do not need to manage Vapi or Retell directly. Devs who want raw provider control can still use Vapi or Retell directly and build their own platform.
Running an agency on Vapi or Retell and outgrowing them?
Hermes is the agency operating layer. One platform. Your brand. From $149 per month.
Apply to the Founders' Beta