ToolColumnAI plans and limits, checked

Buyer guides

AI Voice API Pricing Explained: Characters, Minutes, Credits, and App Boundaries

Understand AI voice API pricing: compare app subscriptions against raw developer APIs, evaluate character vs audio duration metering, and model production scale economics.

Updated

In short

Most text-to-speech APIs bill per character of input text, so estimate characters first: at roughly 750 to 1,000 characters per minute of speech, an hour of audio is 45,000 to 60,000 characters. Two other meters apply to newer products: OpenAI's current speech models and Google's Gemini-TTS bill in tokens, and most managed voice agents bill per minute of conversation, though Azure's Voice Live bills tokens. App subscriptions are a separate wallet: ElevenLabs bills its API in dollars, not the credits its app plans use, and Speechify bills its developer API separately from its Reader app.

On this page4

Standard voices cost $4 per million characters; Google's Studio voices $160

Voice API

Per 1M characters

About per audio hour

Google WaveNet or Standard; Amazon Polly Standard

$4

$0.19

Murf Falcon

$10

$0.48

Azure Neural

$15

$0.72

Amazon Polly Neural; Google Neural2

$16

$0.77

Azure Neural HD

$22

$1.06

Deepgram Aura-2; Google Chirp 3: HD; Polly Generative; Murf Gen2

$30

$1.44

ElevenLabs Flash, Turbo, v4 Turbo, v3 Conversational

$40

$1.92

ElevenLabs v4, v3, v2 Multilingual

$80

$3.84

Google Studio voices

$160

$7.68

Prices are each vendor's pay-as-you-go list rates on October 16, 2026 (Azure for East US). The hour column assumes 48,000 characters per hour, about 800 a minute, which is our assumption for illustration; Unreal Speech uses 750 and ElevenLabs about 1,000. The spread of roughly 40 to 1 between the cheapest standard voices and premium studio voices matters more than small differences inside a tier.

Subscription APIs buy those characters in bulk. Speechify's API includes 1.9 million characters for $10 and charges $10 to $6 per million beyond its plans' allowances; Unreal Speech's plans work out to $16 down to $8 per million; Cartesia's credits are characters too, and dividing each plan's fee by its credits gives about $37 to $50 per million. OpenAI deprecated its tts-1 and tts-1-hd models ($15 and $30 per million characters) on October 1, 2026, with removal from the API set for January 6, 2027.

Token-billed voices need a conversion

Gemini-TTS states its own rate: 25 audio tokens per second, so an hour is 90,000 output tokens. At the preview Gemini 3.8 Flash TTS model's $9 per million audio tokens, that is about $0.81 an hour until December 31, 2026, and about $1.62 from January 1, 2027, when the price doubles, by our arithmetic. OpenAI's replacement for its deprecated speech models is gpt-realtime-2.1-mini at $20 per million audio output tokens, but OpenAI publishes no tokens-per-second rate, so measure a sample before budgeting.

Agent minutes cost $0.05 to $0.163, often before the language model

For live conversations, the meter becomes connection or call time: ElevenLabs agents cost $0.08 a minute, Cartesia's managed agents $0.06, Deepgram's Voice Agent API $0.050 to $0.163 on pay-as-you-go, and OpenAI's GPT-Live $0.05, billed per second. ElevenLabs and OpenAI bill the language model separately, and ElevenLabs passes telephony through at cost; Cartesia adds $0.014 a minute for its phone numbers, and Deepgram's lower rates apply when you bring your own voice, or both your own language model and voice. So the voice rate is a floor, not the call cost.

Free allowances run from 10,000 to 5 million characters a month

Google gives 4 million free characters a month on WaveNet and Standard voices and 1 million on most others, Polly 5 million standard characters a month (newer voice types for 12 months only), Azure 500,000, Speechify's API 500,000, Unreal Speech 250,000, Cartesia 20,000 and ElevenLabs 10,000 to 20,000 depending on the model (both non-commercial); Deepgram gives a $200 credit and Murf $10. OpenAI publishes no free speech allowance. At the limit, behavior differs: Speechify's free API pauses until next month, Cartesia's paid plans block requests unless you opt in to overage, and Unreal Speech's paid plans charge overage daily at the plan's rate.

Plans and prices

Tools in this guide

  • ElevenLabs
    Free tier
    Free10k credits/mo; about 10 TTS UI minutes; 3 Studio projects
    Cheapest paid plan
    $6/moStarter30k credits/mo; about 30 TTS UI minutes; 20 Studio projects
    Top plan
    $99/moPro
    Last checked
    Prices checked
  • Cartesia
    Free tier
    Free20K credits/mo; ~27 TTS min; ~1h51m STT; 1 phone number; $1 prepaid agents
    Cheapest paid plan
    $5/mo + usageProSee plans on the tool page
    Top plan
    Same as cheapest
    Last checked
    Prices checked
  • Speechify
    Free tier
    FreeBasic TTS; 10 robotic voices; up to 1.5x speed
    Cheapest paid plan
    $19/moStudio Starter7,200 Studio credits/mo (1 credit per second of voiceover, about 2 hours); stock assets; commercial usage rights
    Top plan
    $49/moStudio Creator
    Last checked
    Prices checked
  • Unreal Speech
    Free tier
    Free250K characters/mo; about 6 hours of audio; commercial use with a link to unrealspeech.com
    Cheapest paid plan
    Usage-based API
    Top plan
    —
    Last checked
    Prices checked
  • Murf AI
    Free tier
    Free10 mins voice generation; 10 projects; 1 editor; no downloads or commercial rights
    Cheapest paid plan
    $29/mo + usageCreator24 hrs voice generation/year billed annually; 100 projects; 1 editor
    Top plan
    Same as cheapest
    Last checked
    Prices checked
  • Gemini
    Free tier
    FreeStandard compute-based limits that refresh every 5 hours up to a weekly cap; no fixed daily count
    Cheapest paid plan
    $4.99/moGoogle AI Plus2x higher usage access than Free; 200 Google Flow credits; 400 GB storage
    Top plan
    $199.99/moGoogle AI Ultra 20x
    Last checked
    Prices checked

Checked in the last 7 days

Edited byRick Finch

Official sources

Editorial guidance grounded in official product sources.

Show all 19 sources

FAQ

Common questions

Is AI voice API pricing included in a normal app subscription?

Usually no. App subscriptions often cover a hosted studio, creator workflow, reader app, or workspace allowance. API calls can be billed separately by characters, bytes, minutes, seconds, requests, credits, agent time, or negotiated enterprise usage.

Which voice API pricing unit should developers estimate first?

Start with the unit that grows with the product workflow. Scripted TTS usually starts with characters or bytes, speech-to-text starts with audio duration, dubbing starts with source minutes and target languages, and voice agents start with connected minutes plus concurrency.

Are credits comparable across ElevenLabs, Cartesia, Fish Audio, Resemble, or Speechify?

No. Credits are vendor-specific. They can help compare tiers inside one vendor, but they should not be treated as a common currency unless the vendor publishes exactly how credits convert into characters, seconds, minutes, cloning, or agent usage.

When do agent minutes matter more than text-to-speech characters?

Agent minutes matter when the product runs live conversations, phone calls, or real-time support workflows. In that case, session length, idle time, transfers, telephony, and concurrent calls can drive spend more than the text that the agent speaks.

When should a buyer ask for enterprise voice pricing?

Ask for enterprise pricing when public self-serve plans do not cover the needed volume, concurrency, security review, data retention, SSO, support SLA, custom voice rights, on-premise deployment, procurement terms, or predictable overage rules.

Next steps

Take the next evaluation step