Money
Voice AI Pricing Compared: What ElevenLabs, Hume and OpenAI Realtime Actually Cost
Disclosure: some links in this article may earn us a commission at no cost to you. Prices below are the providers' own published rates, checked November 9, 2026, and they change — confirm on the provider's pricing page before you commit. Dessa itself has no price: Block has not announced one.
Dessa AI's price is the easiest line in this comparison: there isn't one. Block has announced no pricing, no API and no paid tier for Dessa, as our Dessa price explainer records. But if you need working voice AI today — and the readers finding our Dessa alternatives guide clearly do — the real question is what the available systems actually cost. The three most compared options price three different ways, and the headline number is never the whole bill.
ElevenLabs: per agent-minute, on a subscription ladder
ElevenLabs sells its conversational product, ElevenAgents, through subscription plans that bundle call minutes. On the published ladder, the Free plan includes 15 agent call minutes a month; Starter is $6 a month with 75 minutes; Creator is $22 with 275; Pro is $99 with 1,238; and Business is $990 with 12,375. The overage rate is the simple part: additional call minutes cost $0.08 per minute on every plan, with burst minutes above your concurrency limit at $0.16. The less visible part is that the language model behind the agent is billed on top, along with any telephony charges — so your true per-minute cost is $0.08 plus LLM and phone costs. For pure text-to-speech rather than agents, ElevenLabs prices by characters through plan credits, a different meter entirely. We compared the technology itself in Dessa vs ElevenLabs.
Hume EVI: the cheapest entry point, emotion included
Hume's Empathic Voice Interface — the emotion-aware system we compared in Dessa vs Hume EVI — has the lowest entry price of the three. Its Free plan includes 5 EVI minutes a month; Starter is $3 a month with 40 minutes; Creator is $14 with 200; Pro is $70 with 1,200; Scale is $200 with 5,000; and Business is $500 with 12,500. Overage minutes get cheaper as you climb: $0.07 per minute on lower tiers down to $0.04 on Business. Hume also sells its Octave text-to-speech by the character on the same plans. As with ElevenLabs, the underlying language model is a separate cost to budget for.
OpenAI: token pricing, plus a flat per-minute option
OpenAI prices its Realtime API by tokens, not minutes, which makes comparison arithmetic necessary. The published audio rates for gpt-realtime-2.1 are $32 per million audio input tokens and $64 per million audio output tokens, with cached audio input at $0.40 per million; the mini model is $10 in and $20 out. Independent cost guides that model a typical two-way conversation from those rates land around $0.05–$0.15 per minute on the flagship with caching working, and roughly $0.02–$0.05 on the mini. Separately, OpenAI's GPT-Live — the model Block's own demo page names as GPT-Live-1 — is offered at a flat $0.05 per session minute plus the cost of the backend model it delegates to. That flat rate is the easiest OpenAI number to budget, and the one most directly comparable to the per-minute prices above.
How to compare honestly
Three rules keep this comparison fair. First, compare the same unit: a conversation minute, including who pays for the language model and the phone line. ElevenLabs' $0.08 and Hume's $0.04–$0.07 overage both exclude pieces OpenAI bundles differently. Second, compare at your volume: at 100 minutes a month you live on entry plans; at 10,000 minutes, overage rates and concurrency limits decide the bill. Third, remember what none of these prices buy you: none of these systems is Dessa. Each is a text-rooted pipeline or a model built on a text foundation, available and supported today; Dessa's speech-only approach remains a research preview you cannot buy at any price.
The bottom line
For a first experiment, Hume's $3 Starter or ElevenLabs' free agent minutes cost almost nothing to try. For a production phone agent, model your real call volume against ElevenLabs' $0.08 overage, Hume's tiered overage and OpenAI's token rates before choosing. And for Dessa, the correct budget line is still zero — not because it is free, but because Block has not announced a price, an API or a date. When that changes, this page will be updated the same day.
Sources: ElevenLabs, Hume and OpenAI published pricing pages and rate cards as checked November 9, 2026 (ElevenAgents plan minutes and $0.08/min overage; Hume EVI plans and $0.07–$0.04/min overage; OpenAI Realtime token rates and GPT-Live $0.05/session minute); Block's official Dessa page — no Dessa pricing announced.
Frequently asked questions
How much does voice AI cost per minute?
On published rates checked November 9, 2026: ElevenLabs Agents charge $0.08 per extra call minute plus LLM and telephony; Hume EVI overage runs $0.07 down to $0.04 per minute by tier; OpenAI's GPT-Live is $0.05 per session minute plus its backend model, while its Realtime API bills by tokens.
What does Dessa AI cost compared to these?
Nothing comparable exists: Block has announced no Dessa pricing, API or paid tier. The prices in this article are for rival systems you can actually use today, not for Dessa.
Which voice AI is cheapest to start with?
Hume's Starter plan is $3 a month with 40 EVI minutes, and ElevenLabs offers a free tier with 15 agent minutes. OpenAI's realtime models bill by usage with no subscription, so a small test can also cost very little.
🚀 Get Dessa launch alerts
Dessa is still a research preview. Join the free DessaAI Daily list and we will email you the day access opens — plus the week's biggest Dessa stories, once a week. No spam, no hype.
Form not loading? Subscribe on the signup page →