AI voice agent cost calculator: what a minute really costs

On the developer platforms I priced on Oct 10, 2026, an inbound minute costs about $0.090 to $0.155 all in, once speech to text, the language model, the voice and the phone line are added (Vapi, Retell AI and ElevenLabs pricing pages, plus Twilio's United States rates). Bland's flat Start rate comes to $0.149 with the carrier. The headline rates vendors advertise usually cover only part of that minute.

Vapi's $0.05 is the hosting fee. Speech to text, the model, the voice and transport are each billed at the provider's price on top of it (Vapi pricing page, checked Oct 10, 2026). With the cheapest options in Vapi's own calculator plus a Twilio number, the minute comes to $0.090, which is 1.8 times the headline. With the top of each range it is $0.137.

Often, yes. Vapi charges $2,000 a month per organization for HIPAA eligible data handling with a BAA (Vapi pricing page and HIPAA docs). Retell AI lets you self sign a BAA with no additional fee (Retell compliance docs). At 5,000 minutes a month that puts Vapi at $0.491 a minute against $0.093 for Retell on the default stacks, and Vapi's usage only outgrows the flat fee at about 21,815 minutes a month. For Bland, ElevenLabs Agents and OpenAI direct I found no HIPAA price on the pages I checked, so the calculator flags them rather than guessing.

Concurrent calls are how many callers are on the line with the agent at the same moment. The calculator estimates it from your busiest hour with the Erlang B formula, sized so a caller has at most a 1% chance of finding every line busy. Vapi includes 4 lines, its $29 Core package raises that to 10, and extra lines are $10 a month each. Retell AI includes 20 and charges $8 a line beyond that. Bland's Start plan caps you at 10 at once and 100 calls a day (pricing pages, checked Oct 10, 2026).

Bland's per minute rate includes the model, transcription and voice, with no token charges (Bland pricing page). The phone line is billed separately, at pass through cost on Bland's own Twilio or directly by your carrier (Bland billing docs). Start is $0.14 a minute with no fee and Build is $0.12 plus $299 a month, so Build only becomes cheaper above 14,950 minutes a month. Against the component platforms, the bundle usually costs more per minute at the default stacks.

The BLS median pay for receptionists was $38,010 in May 2025. Wages are 68.7% of total compensation for private industry office and administrative support workers (BLS Employer Costs for Employee Compensation, June 2026), so the loaded cost is about $55,328 a year, or $4,611 a month. Even if that person spent every paid minute on the phone, that is $0.44 a minute, for 40 hours a week and one call at a time.

PATLive's published plans run from $759 for 500 minutes to $6,899 for 5,000, with extra minutes billed at $1.49 to $2.99. Ruby charges $250 for 50 minutes up to $1,725 for 500 (pricing pages, checked Oct 10, 2026). Per included minute that is $1.38 to $5.00, roughly 9 to 55 times the AI platform rates above, though a person answers every call.

You can. GPT-Live 1 costs $0.05 a voice minute billed per second, with the backend model billed separately (OpenAI model page). gpt-realtime-2.1 is priced per audio token: $32 per million in, $64 out and $0.40 cached, where a minute of caller speech is 600 tokens and a minute of agent speech is 1,200 (OpenAI pricing page and voice cost guide). Every turn sends the whole conversation so far again, so longer calls cost more per minute and caching, which OpenAI calls best effort, matters. On the example inputs that is about $0.066 a minute for GPT-Live 1 and $0.071 for gpt-realtime-2.1 with a Twilio line, but you build the call handling, tools, logging and phone connection yourself.