Cost Comparator · Voice agent

What does an AI voice agent cost per call?

Voice is three billed legs per turn: transcription, the model, and speech synthesis. Most estimates only price one of them.

Model my bill

A voice agent bills differently than a chat product. Every turn of a phone call runs through speech-to-text (per minute), the language model (per token, on a much shorter exchange than chat), and text-to-speech (per minute, again). Teams that estimate voice cost by only pricing the LLM leg routinely under-forecast by 3 to 5x, because the STT and TTS legs together often cost more than the model turn itself.

Call duration is the real cost driver, not message count. A 4-minute support call bills roughly 4 minutes of transcription and 4 minutes of synthesis regardless of how many back-and-forth turns happen inside it, so the number to nail down first is average call length, not average messages per call.

Who runs this workload

Outbound and inbound call centers replacing or augmenting human agents
IVR replacements for appointment booking, order status, or basic triage
Voice-first products (in-car assistants, accessibility tools, phone-based onboarding)

Cost drivers

What actually moves the bill.

Call duration

STT and TTS both bill per minute. A 30-second call and a 5-minute call are a 10x cost difference regardless of turn count.

Latency-tier model

Voice needs fast time-to-first-token. That usually means a smaller, faster model, which is also the cheaper one, a rare case where speed and cost point the same direction.

Interruption handling

Barge-in and re-prompting add extra STT and LLM calls per turn if not handled carefully in the agent loop.

FAQ

Common questions.

How much does an AI voice agent cost per minute?

Stacking a standard STT provider (roughly $0.004 to $0.017/minute), a fast LLM turn, and a standard TTS voice (roughly $0.015 to $0.16/minute depending on quality) typically lands between $0.05 and $0.35 per minute of call time, before your own telephony carrier costs.

Why is my voice agent bill higher than my chatbot bill?

Because voice pays for two audio legs (transcription and synthesis) on top of the model call that a text chatbot never pays. A voice minute usually costs several times more than an equivalent chat exchange of the same length.

What is the cheapest way to build a voice agent?

Pair a fast, cheap LLM (Claude Haiku 4.5 or GPT-4.1 mini) with a standard-tier TTS voice instead of a premium one, and keep call scripts tight so average duration stays low. Premium voice quality is the single biggest lever if you need to cut cost.

Other workloads