Vapi vs Retell vs Bland in 2026: What an AI Voice Agent Really Costs Per Minute, and Which One Wins at 20,000 Calls

TL;DR: Checked September 4, 2026: Vapi charges $0.05 per minute for hosting and passes every model through at cost, which lands a realistic inbound stack at about $0.105 per minute. Retell bundles its own per-minute LLM and voice rates and lands at $0.11 to $0.135. Bland charges a flat $0.14 (Start), $0.12 (Build, $299 per month) or $0.11 (Scale, $499 per month) plus telephony. At 1,000 minutes Vapi and Retell are within $6 of each other. At 20,000 minutes Vapi costs $2,101, Retell $2,202, and Bland $2,869 because the Start plan’s 100 calls per day cap forces you onto a paid tier.

The three pricing models, in one table

The headline per-minute numbers are not comparable. Vapi’s $0.05 is hosting only. Retell’s $0.055 is voice infrastructure only. Bland’s $0.14 includes the LLM, transcription and voice but not the phone line. Here is what each vendor actually lists.

Sources: Vapi pricing, Vapi phone calling docs, Retell pricing, Bland pricing. All checked September 4, 2026.

Bland also lists a transfer-time rate of $0.05, $0.04 and $0.03 per minute by tier, waived if you bring your own telephony.

What the components cost when you buy them yourself

Vapi’s “at cost” model means you pay the underlying vendors. These are the list prices on September 4, 2026.

Two estimates drive the last column. First, the agent speaks about half of every minute, so TTS burns roughly 500 characters per call minute rather than 1,000. If your agent monologues, double the TTS line. Second, a voice agent makes about six LLM calls per minute, each carrying roughly 2,000 input tokens (system prompt plus growing transcript) and returning about 60 output tokens: 12,000 input and 360 output tokens per minute.

The Haiku 4.5 arithmetic: 12,000 / 1,000,000 x $1 = $0.012 input, 360 / 1,000,000 x $5 = $0.0018 output, total $0.0138. Sonnet 5 at $2 / $10 gives $0.024 + $0.0036 = $0.0276. Prompt caching cuts this further: at $0.10 per 1M cached input on Haiku, an 80 percent cache hit rate drops the input side to about $0.0034, so the whole LLM line falls near $0.005 per minute. For the full model menu see the AI API pricing September 2026 tracker and the Anthropic API pricing breakdown.

All-in cost per minute, platform by platform

Stack the components and you get real per-minute numbers. Inbound local calls, Deepgram Nova-3 at the standard rate, ElevenLabs Flash where named.

Vapi with your own keys, Haiku 4.5: $0.05 hosting + $0.0077 STT + $0.025 TTS + $0.0138 LLM + $0.0085 Twilio = $0.105 per minute.

Vapi with Sonnet 5: swap the LLM line, $0.105 – $0.0138 + $0.0276 = $0.119 per minute.

Retell, cheapest sensible stack: $0.055 voice infrastructure + $0.015 Retell platform voice + $0.025 Claude 4.5 Haiku + $0.015 Retell telephony = $0.11 per minute.

Retell with ElevenLabs voice: $0.055 + $0.040 + $0.025 + $0.015 = $0.135 per minute. Move to Claude 5 Sonnet and it is $0.19.

Bland Start: $0.14 + $0.0085 Twilio pass-through = $0.1485 per minute, no monthly fee.

Bland Build: $0.12 + $0.0085 = $0.1285 per minute plus $299 per month. Bland Scale: $0.11 + $0.0085 = $0.1185 per minute plus $499 per month.

Retell’s LLM line deserves a second look. Its Claude 4.5 Haiku line is $0.025 per minute; the same model bought direct comes to about $0.0138 on the assumptions above. Claude 5 Sonnet is $0.08 on Retell against roughly $0.0276 direct. That spread is Retell’s margin for handling the model integration, and it is the single biggest reason Vapi with your own keys ends up cheaper at every volume.

What you actually pay at 1,000, 5,000 and 20,000 minutes

Assume 3-minute average calls, inbound local numbers, and the cheapest credible stack on each platform (Vapi with Haiku 4.5, Retell with its own platform voice and Haiku, Bland on the lowest plan its caps allow). Numbers are monthly.

The Bland jump at 20,000 minutes is a cap problem, not a rate problem. Start allows 100 calls per day. At 222 calls per day you must move to Build (2,000 per day) or Scale (5,000 per day). Build is $299 + $2,570 = $2,869. Scale is $499 + 20,000 x $0.1185 = $2,869. Identical to the dollar, so Scale only pays off above 20,000 minutes.

Effective all-in cost per minute at 20,000 minutes: Vapi $0.105, Retell $0.110, Bland $0.143. ElevenLabs voices on Retell add $0.025 per minute: $137, $677 and $2,702. Sonnet 5 on Vapi adds $0.0138: $119.15, $596.15 and $2,377.15.

Hidden costs that move the bill

Concurrency. Vapi includes 10 concurrent lines and charges $10 per line per month above that. Retell includes 20 and charges $8 per extra concurrent call per month. Bland caps at 10 on Start, 50 on Build, 100 on Scale, with no add-on price, only the next plan. At 20,000 minutes across a 10-hour business day, average concurrency is roughly 1.1 calls (667 minutes per day / 600 minutes); a peak of five times average is still under every floor. Concurrency only bites on outbound campaigns that fire hundreds of dials at once.

Daily and hourly caps. Only Bland has them: 100 calls per day and per hour on Start, 2,000 per day and 1,000 per hour on Build, 5,000 per day and 1,000 per hour on Scale. An outbound blast hits the hourly cap first.

Number rental. Vapi gives up to 5 free US numbers, but outbound calling is not supported on them, so any outbound use case means a Twilio number at $1.15 per month plus $0.014 per outbound minute. Retell numbers are $2 per month, a verified number is $10 per month. Twilio toll-free inbound is $0.022 per minute, nearly three times local.

Model markup. Covered above: Retell’s per-minute LLM rates run about 1.8x to 2.9x the direct API cost on the stated assumptions. Retell’s Fast Tier doubles them again (GPT 5.5 at $0.32 per minute). Bland hides the LLM entirely and does not let you pick the model, which is fine until you need a specific one.

Add-ons. Retell charges $0.10 per outbound branded call, $0.005 per minute for denoising, $0.01 per minute for PII removal and $0.10 per minute for AI quality assurance, which nearly doubles a $0.11 stack. Vapi’s HIPAA add-on is $2,000 per month and zero data retention is $1,000 per month. Bland puts BAA, SSO and data residency on Enterprise only.

Promo cliffs. Deepgram lists Nova-3 at $0.0048 as a limited-time streaming promo against $0.0077 standard, and Flux TTS is free until September 12, 2026, then $0.045 per 1,000 characters. ElevenLabs is advertising 50 percent off TTS API pricing for subscriptions before September 11, 2026. Budget on standard rates.

Retention. Vapi’s Build tier keeps call history for 14 days.

What operators charge to set this up

One public marketplace snapshot from September 4, 2026, the PeoplePerHour AI voice agent category. Listed fixed-price offers include “build an AI voice agent using Bland AI” at $140, “build an AI voice agent using Vapi” at $250, “build an AI voice agent using Retell AI” at $275 and $350, “AI Voice Receptionist for Small Businesses” at $475, “build a custom AI voice agent using Retell AI and GoHighLevel” at $690, and two 24/7 receptionist builds at $1,500. The page spans $50 to $1,510. Fiverr and Upwork listing prices are not quoted here.

The margin math for an operator running a client’s agent is straightforward. A 5,000-minute client costs $526 on Vapi. Resell at double cost, $0.21 per minute (an assumption, not a market rate), and the client pays $1,050, leaving $524 per month of recurring margin per client on top of the build fee. For where build fees sit against other automation work, see the AI automation rate card.

BetOnAI Verdict

Under 1,000 minutes, pick on convenience, not price. Vapi and Retell are $6 apart. Retell’s per-minute LLM pricing means no API keys to manage and a bill that is easy to explain to a client. Bland at $148.50 costs 40 percent more but includes the number and hides every provider decision. A non-developer launching one receptionist gets the least work per dollar on Bland Start.

At 5,000 minutes, Vapi with your own keys wins and Retell is close. $526 versus $552. The $26 gap is smaller than one hour of your time, so choose Retell if you would rather not manage Deepgram, ElevenLabs and Anthropic accounts. Bland Start at $742.50 is $216 more than Vapi every month with nothing on the Start plan to justify it.

At 20,000 minutes, Vapi wins and Bland loses on caps. Vapi $2,101, Retell $2,202, Bland $2,869. Bland’s $768 premium buys no model choice and daily call limits. Only pay it if the transfer, pathway and guardrail features are the product.

What to do this week. One: run inbound agents on local numbers, never toll-free, because Twilio’s $0.022 inbound rate adds $270 per month at 20,000 minutes for nothing. Two: on Vapi, cache the system prompt; on the stated assumptions that drops the LLM line from $0.0138 to about $0.005 per minute, $176 per month at 20,000 minutes. Three: when quoting an agency client, price at 2x your Vapi cost and stay on Haiku 4.5 or Gemini 3.5 Flash-Lite unless call quality demands Sonnet 5. The AI Pricing Watch hub tracks the model prices those numbers depend on.

Frequently Asked Questions

How much does an AI voice agent cost per minute in 2026?

Between $0.105 and $0.19 per minute all-in on the three main platforms, checked September 4, 2026. Vapi with your own model keys lands at about $0.105, Retell at $0.11 to $0.135 depending on the voice, and Bland at $0.11 to $0.14 plus telephony, with a monthly fee on its Build and Scale plans.

Which is cheaper, Vapi, Retell or Bland?

Vapi is cheapest at every volume when you bring your own keys: $106 at 1,000 minutes, $526 at 5,000 and $2,101 at 20,000. Retell is within $6 to $100 of it. Bland costs 40% more at low volume and $768 more at 20,000 minutes because its Start plan caps calls at 100 a day.

What hidden costs should I budget for with AI voice agents?

Concurrency add-ons ($8 to $10 per extra line per month), number rental, toll-free inbound at nearly three times the local rate, platform markups on the LLM (Retell charges 1.8x to 2.9x the direct API price), and add-ons like PII removal or quality assurance that can double a per-minute rate.

Sources

• Vapi, pricing page: https://vapi.ai/pricing
• Vapi, phone calling docs: https://docs.vapi.ai/phone-calling
• Retell AI, pricing page: https://www.retellai.com/pricing
• Bland AI, pricing page: https://www.bland.ai/pricing
• Twilio, Programmable Voice pricing, United States: https://www.twilio.com/en-us/voice/pricing/us
• ElevenLabs, API pricing: https://elevenlabs.io/pricing/api
• Deepgram, pricing: https://deepgram.com/pricing
• Anthropic, Claude pricing: https://platform.claude.com/docs/en/about-claude/pricing
• OpenAI, API pricing: https://developers.openai.com/api/docs/pricing
• Google, Gemini API pricing: https://ai.google.dev/gemini-api/docs/pricing
• PeoplePerHour, AI voice agent freelancers: https://www.peopleperhour.com/hire-freelancers/ai-voice-agent