How Much Do AI Voice Agents Cost? Per-Minute Pricing Decoded

A working AI voice agent costs roughly $0.07 to $0.31 per call minute on pay-as-you-go platforms in 2026, or $0.12 to $0.14 per minute on Bland's all-in plans, or about $1.67 to $2.00 per call on Smith.ai's per-call AI receptionist. That is the honest ballpark. The reason quotes for the same phone line come back at wildly different numbers is that "per minute" is not one price - it is a stack of meters, some visible and some buried, plus fixed fees that only matter at your volume.

This guide decodes the stack using published rates we verified in September 2026 from Vapi, Retell, Bland, ElevenLabs, Deepgram, Smith.ai and Synthflow. Every number below traces to a vendor pricing page. None of it is a quote; provider pricing changes and your configuration decides your rate.

Key takeaways

• A per-minute price hides up to five meters: platform hosting, speech recognition, the language model, text-to-speech and telephony.
• Verified platform rates: Vapi charges $0.05 per minute in hosting; Retell's pay-as-you-go range is $0.07 to $0.31 per minute; Bland is $0.14 per minute on its free-tier plan and $0.12 with a $299 monthly platform fee.
• All-in-one options shift the shape of the bill: ElevenLabs Agents charges $0.08 per additional call minute, and Smith.ai charges per call from $1.67 at 300 calls per month.
• Model choice moves cost more than anything else. Retell's public table spans roughly $0.003 to $0.38 per minute for language models alone, and voices range from $0.015 to $0.040 per minute.
• Fixed fees are where small volumes get crushed: Vapi's HIPAA add-on is $2,000 per month, Synthflow's enterprise contracts start at $30,000 per year, and concurrency beyond 20 calls costs $8 per line per month on Retell.
• Convert every quote to cost per answered call at your volume, then add the fixed fees. That number, not the headline rate, is the one to compare.

The five meters behind one call

When a vendor says "$0.07 per minute," check which meters are inside. Retell's $0.07 lower bound is the platform layer before models, voices and telephony with your cheapest configuration; its own calculator defaults to $0.11 per minute all-in. Bland, by contrast, bundles LLM, speech recognition and text-to-speech into its $0.12 to $0.14 rate and bills telephony separately at pass-through cost.

What the major platforms actually charge

Two things stand out from that table. First, the platforms built for developers (Vapi, Retell, Bland) compete on cents per minute, and the platforms built for buyers (Smith.ai) compete on predictability per call. Second, every cheap per-minute rate comes with a fixed fee somewhere - a monthly package, a platform fee or an enterprise floor - so the effective rate depends on volume more than on the price list.

A worked 1,000-minute month

Vapi's own pricing calculator, run at 1,000 minutes per month in September 2026, produced this estimate:

That is the vendor's own arithmetic, not ours, and it shows the honest shape of a platform bill: the hosting fee is only about half the cost, and model choice swings the total by nearly 60 percent. Swap in a heavier realtime model or a premium voice and the same 1,000 minutes costs meaningfully more.

As a second cross-check, adding published component prices from different vendors - Deepgram's promotional streaming recognition at $0.0065 per minute, a mid-range language model around $0.04 per minute, a platform voice at $0.015 per minute and carrier fees of $0.008 to $0.014 per minute - lands near $0.07 per minute before any engineering time. The stack is cheap. Building and maintaining it is the expensive part.

What actually moves your bill

• Call length, including silence. Retell's billing FAQ states that billing covers the entire call duration because the recognition engine keeps listening, even during silence. A polite agent that waits through long pauses costs more than a brisk one.
• Model choice. Retell's public table spans from about $0.0032 per minute for its lightest model to $0.38 per minute for realtime speech-to-speech. This single decision can multiply your bill by more than 100x.
• Voice choice. Platform voices run $0.015 per minute against $0.040 for premium third-party voices on Retell. On a 2,000-minute month, that difference is $50.
• Add-ons. Retell prices knowledge base access and advanced denoising at $0.005 per minute each, and PII removal at $0.01 per minute. Small per-minute numbers become real money at scale.
• Concurrency and bursts. Calls beyond 20 simultaneous lines cost $8 per line per month on Retell. ElevenLabs charges double - $0.16 per minute - for calls that overflow your plan's concurrency limit, which is a quiet way for a busy Monday to cost more.
• Transfers. Bland bills transfer time at $0.05 per minute on its entry plan and $0.04 on Build. If your agent's main job is routing, that meter matters.
• Compliance. Vapi prices HIPAA-eligible handling, including a signed BAA, at $2,000 per month. Healthcare work is not a per-minute decision.
• Minimum commitments. Bland Build requires a $299 monthly platform fee, Vapi Pro carries a $999 monthly minimum, and Synthflow's enterprise contracts start at $30,000 per year. A low per-minute rate behind a high floor is not a low-cost option at small volume.
• What is not billed. Retell does not charge for calls that fail to connect, and voicemail billing applies only while the agent is active on the line. Ask every vendor what counts as billable, and get it in writing.

The only comparison that matters

Convert every quote into cost per answered call:

(monthly platform fees + usage + expected overage) divided by the number of calls you expect to answer.

As an illustration using Retell's default calculator rate of $0.11 per minute: a three-minute average call costs about $0.33 to answer. If roughly 30 percent of answered calls book a job, that is about $1.10 per booked job in voice costs - before you count the value of the jobs themselves. That arithmetic is ours, built on published rates and assumptions you should replace with your own average call length and booking rate.

Run the same conversion on each quote, add the fixed fees you will actually pay, and rank by cost per answered call. When the numbers are close, decide on the things pricing tables do not show: who owns the phone number, what data you can export, how changes are priced, and how fast you can get a human on the phone when the agent is stuck.

When project pricing fits better

Metered pricing suits simple, stable answering needs. Project pricing suits businesses with specific call flows, CRM requirements and a preference for ownership. Praktivo builds voice systems as projects - Workflow Sprints from $1,500, Full Journey Builds from $5,000, and an optional management plan from $300 per month - with the workflows and data owned by you, not rented per minute. The trade-off is real: it is a build, not a subscription you can switch off after a quiet month. Details are on the pricing page, and the AI voice agent service shows how the workflows are scoped.

If you are comparing vendors more broadly, the AI receptionist pricing guide covers the flat, per-call and per-resolution models side by side, the law firm answering cost breakdown shows how the math looks in one industry, and the AI voice receptionist workflow is the reference build these rates are measured against. The platform-level detail lives in the voice agent guide.

A checklist before you sign

• Write down your monthly call volume and average call length.
• Convert each quote to cost per answered call using the formula above.
• Add the fixed fees: platform packages, concurrency, numbers, compliance.
• Ask which meters are included and which are pass-through.
• Ask what counts as a billable call, and how spam, wrong numbers and voicemail are handled.
• Confirm the overage rate, the burst rate and the minimum commitment in writing.
• Check who owns the phone number, the prompts and the call data.
• Re-run the arithmetic at double your volume. The cheapest option at 200 minutes is rarely the cheapest at 2,000.

FAQ

What is the average cost per minute of an AI voice agent?

Verified pay-as-you-go rates in 2026 run from about $0.07 to $0.31 per minute on Retell, and $0.12 to $0.14 per minute on Bland's all-in plans. ElevenLabs Agents charges $0.08 per extra call minute before language model costs. Your effective rate depends on the models, voices and add-ons you choose, not just the headline.

Why is per-minute pricing split into so many components?

One call consumes speech recognition, a language model, text-to-speech and telephony. Some platforms bill each at provider cost or with markup, and some bundle everything into one rate. Vapi, for example, charges $0.05 per minute in platform hosting and passes model costs through, which is why its estimates show four separate lines.

Is per-call pricing cheaper than per-minute pricing?

It depends on call length. Smith.ai's AI receptionist is $150 per month for 75 calls, about $2 per call, with $2.50 overage. Paid minutes usually cost cents, so per-minute wins on short calls and loses on long ones unless the agent resolves them. Per-call plans cap the meter, which some owners value more than the lowest unit price.

What hidden costs should we expect?

Phone numbers, concurrency limits, HIPAA handling, transfer minutes, burst pricing and minimum commitments. Vapi's HIPAA add-on is $2,000 per month, Bland's Build plan carries a $299 monthly platform fee, and ElevenLabs doubles the per-minute rate for calls that run over your concurrency limit.

Is it cheaper to build on a platform or buy a managed service?

Platforms are cheaper per minute but you own prompt engineering, testing, monitoring and integrations. Managed per-call services bundle that labor into the price. The fair comparison adds your team's time to the platform column, or checks whether your call flows are simple enough to run without it.

Next step

If you want your own numbers rather than a ballpark, bring your call volume and average call length to the Praktivo funnel, or book a call and we will convert them into cost per answered call for your specific call types. We price builds as projects, not minutes, so the comparison is straightforward - see how we scope voice agents and the pricing page for what that looks like.