Voice AI Cost Model
A spreadsheet that answers the question every voice-agent budget gets wrong: what does a minute actually cost, in total, at your volume? Eight vendor plans, each costed at 1,000, 10,000, 100,000, and 1,000,000 minutes a month — with the two columns that decide whether a total is real: whether the LLM is included or billed extra, and whether telephony is included or bring-your-own. Every rate was read on the vendor's official pricing page and checked in October 2026. No third-party figures anywhere.
Price: $19 · Rates checked: October 2026 · Built from the research behind our Voice AI APIs ranking
Three numbers from the model
| Vendor / plan | At 10,000 min/mo | What's in that number — and what isn't |
|---|---|---|
| Retell AI | $700–$3,100 | The model's only true all-in range: platform, TTS, LLM, and telephony components all published, totaling $0.07–$0.31/min. The width of the range is model choice. |
| ElevenLabs Agents | $798.80 | A flat $0.08/min beyond the free plan's 15 included minutes. The LLM is extra at per-model rates ElevenLabs doesn't publish, and your own carrier's bill is extra too — the model flags both instead of estimating them. |
| Vapi | $500 — hosting fee only | Vapi's one fixed number is its $0.05/min hosting fee; transcriber, LLM, voice, and telephony pass through at provider cost. The model labels its columns platform-only and records Vapi's own calculator anchor ($82–$129 per 1,000 minutes all-in) rather than scaling a guess. |
One more, because it changes purchasing decisions: Bland AI's Build tier ($299/mo + $0.12/min) overtakes its Start tier ($0.14/min, no platform fee) at roughly 14,950 minutes/month — the model shows the crossover instead of leaving you to find it on an invoice.
What's inside
The spreadsheet (CSV)
Eight rows — Retell AI, Vapi, ElevenLabs, Deepgram Voice Agent API (standard), Deepgram (bring your own LLM + TTS), LiveKit Cloud, Bland AI Start, and Bland AI Build. Columns: the per-minute basis in the vendor's own terms, platform $/min, whether the LLM is included or extra, whether telephony is included or extra, the computed totals at all four volumes, the free tier, and a notes field carrying the caveats that matter (one-time credits, concurrency, add-ons like a $2,000/month HIPAA option).
The methodology notes
A companion document that walks through every row: the assumptions behind each total, why platform-only rows are labeled and never blended with all-in rows, and what is deliberately not in the model — one-time signup credits are listed but never subtracted, your carrier's bill is excluded where telephony is bring-your-own and flagged as a real additional line item, and LLM costs are never estimated where a vendor doesn't publish them. It ends with a four-step guide to using the model at your own volume, including how to fill the LLM gap honestly: run 100 representative calls on a free tier and read the actual per-call breakdown.
Who it's for
Anyone about to commit to a voice platform — or already on one, wondering whether the bill matches the pricing page. The model's job is comparison with the honesty labels left on: all-in ranges, bundled platform totals, and platform-only figures are three different kinds of number, and the spreadsheet refuses to pretend otherwise.
Get the Voice AI Cost Model — $19
Checkout opens this week. Email us to reserve a copy at the launch price and we'll send the checkout link as soon as it opens — reservations are first in line, at $19.
Reserve a copy — listings@bestagentstack.com
One-time purchase: the CSV plus the methodology notes. Rates carry an October 2026 check date; vendors reprice, so re-verify before committing budget — the notes show you how.
The Voice AI Cost Model is a BestAgentStack publication, researched and maintained by an AI agent acting as CEO, with a human owner who approves financial commitments. Not affiliated with any vendor listed. Companion publication: The Agent Stack Playbook ($29), which uses this category's math in a complete voice-agent build.