Voice AI Cost Calculator
Compare what your call volume costs on managed platforms vs a custom LiveKit stack, using real production pricing, not vendor marketing.
$0.025/minute(2.5¢): our entire production voice stack, all in.
How we calculate this
Your stack = selected STT + selected TTS (at LiveKit's published assumption of 600 characters/minute of agent speech) + selected LLM (at 3,000 input / 175 output tokens/minute) + a fixed $0.0072/min telephony line (Telnyx SIP + LiveKit SIP fee) + a fixed $0.0051/min self-hosted infra line (agent workers, Redis, observability: the gap between list-price and our measured production cost).
Every provider rate traces to a published price: LiveKit's inference rate card, or the provider's own pricing page. See the full line-by-line breakdown in our build-vs-buy math. Competitor rows (Retell, Vapi, Bland) are their own published pricing, not affected by your stack picks. Rates checked Aug 2026. Vendor pricing changes, so treat this as directionally accurate, not a live feed.
Why the gap is usually this large: Retell's own pricing calculator attributes $0.055/min of its $0.11/min default configuration to "voice infrastructure": a platform fee charged on top of the underlying STT, LLM and TTS cost. Vapi's platform fee is $0.05/min, also on top of whatever STT/LLM/TTS you bring. A self-hosted stack has no platform markup, only the real component costs, plus the smaller infra line above.
$0.025/min is our measured steady-state production cost, not a promotional rate, and it's not subsidized by infrastructure credits. The defaults above reproduce it exactly.
Past the crossover point? See our AI voice agent development service: the migration that took production cost from 10¢ to 2.5¢/minute.
Common
questions.
Straight answers. If yours isn't here, ask on a 20-minute call.
How much does voice AI cost per minute in 2026?+
Managed platforms (Retell, Vapi, Bland) run roughly 8 to 15¢ per minute all-in once you add STT, LLM, TTS and telephony. A custom LiveKit stack with optimized providers runs around 2.5 to 4¢ per minute at production quality. The 2.5¢ figure is our own production number.
When is a custom voice stack worth it?+
Around 20,600 minutes per month the savings from owning the stack usually exceed the build cost within one quarter. Below that, managed platforms are the right choice. We say this even though we sell custom builds.
What's included in these per-minute estimates?+
Speech-to-text, LLM inference, text-to-speech, WebRTC/telephony transport and infrastructure. Platform fees are included for managed options; engineering cost is excluded on the custom side (that's the one-time build).
How much does it cost to develop a voice AI agent?+
Our Voice AI agent build runs from $2,800 to $3,600 over 2 to 3 weeks, covering inbound calls around the clock, warm transfer to a person, and recordings, transcripts and QA. Running costs are separate, and they are what this calculator shows. Below about 20,600 minutes a month, a managed platform is usually the cheaper choice overall.
What assumptions does the calculator use per minute of conversation?+
It uses LiveKit's published conversation shape: 600 characters of agent speech a minute for text-to-speech, and 3,000 input plus 175 output tokens a minute for the LLM. On top it adds a fixed $0.0072 a minute for telephony, covering a SIP trunk and LiveKit's SIP fee, and $0.0051 a minute for self-hosted agent workers, Redis and observability.
Which STT, TTS and LLM does Axionry use in production?+
Our production stack runs on LiveKit for WebRTC transport, xAI for streaming speech-to-text and text-to-speech, Gemini 2.5 Flash-Lite as the LLM, and Telnyx SIP for telephony. That combination is the calculator's default, and it reproduces our 2.5¢ a minute steady-state cost. You can swap any layer to see how a different provider changes the total.
Why do managed voice AI platforms cost more per minute?+
Mostly because of the platform fee. Managed platforms charge for their voice infrastructure on top of the underlying speech-to-text, LLM and text-to-speech costs. A self-hosted stack pays only the component costs plus a smaller infrastructure line. You take on the engineering instead, which is why we recommend managed platforms below about 20,600 minutes a month.
Which part of a voice AI stack costs the most?+
Text-to-speech has the widest spread. On the rate card it runs from free to $300 per million characters, so a premium voice can cost more than the rest of the stack combined. The LLM is the next big lever, and fixed telephony and infrastructure add about 1.2¢ a minute. The calculator lets you change each layer and watch the total move.
How can I reduce my voice AI cost per minute?+
Pick the cheapest text-to-speech voice your users accept, a small fast LLM, and streaming speech-to-text priced per minute. Above about 20,600 minutes a month, move from a managed platform to a custom stack. Then look for credits: we secured up to $100K in LiveKit credits, which made eligible voice infrastructure free while they lasted. That path took us from 10¢ to 2.5¢.
How current are the voice AI rates in this calculator?+
Provider rates come from LiveKit's published inference rate card and each provider's own pricing page, checked in August 2026. Managed platform rows use those platforms' own published pricing. Vendors reprice often, so treat the results as directionally accurate rather than a live feed, and confirm the latest rate card before committing to a provider.
Want your stack at 2.5¢/minute?
Twenty minutes, straight to the engineer. No sales rep, no deck.