Direct answer: public AI voice-over prices in 2026 fall into three bands. OpenAI tts-1 is USD 15 per million characters and tts-1-hd is USD 30 per million characters. Google Standard/WaveNet is USD 4 per million characters, Neural2 USD 16, Chirp 3 HD USD 30, and Studio USD 160. ElevenLabs bills by subscription credits: Creator is about USD 22 for 121,000 characters (about USD 182 per million), while Pro/Scale/Business are about USD 165–166 per million. At 1 million characters/month, per-character APIs cost about USD 4–160, while subscriptions cost about USD 165–299; commercial licensing, voice cloning, pronunciation review, video production, and human listening often dominate TCO.
This guide is for marketing, training, customer service, IT, and procurement teams producing product videos, e-learning, IVR prompts, multilingual voice-over, or short-form content. Data cut-off: 31 October 2026. Prices come from official OpenAI, Google Cloud, and ElevenLabs pages. Azure TTS did not provide a stable public unit price during this check and is excluded from the numeric comparison. All prices exclude tax; USD and RMB are not converted and no ranking is implied. This is public-source research and an editorial model, not a quote, and it does not promise voice quality, naturalness, savings, or a go-live date.
A per-character API and a credit subscription are different bills. The subscription premium pays for the voice library, cloning, UI, and commercial licence, not just synthesis. Voice cloning, celebrity voices, and commercial scope need consent and rights before price.
What are the official AI voice-over billing units in 2026?
OpenAI/Google mainly bill per character or token; ElevenLabs bills by subscription credits. Separate characters, tokens, credits, and minutes first.
There are at least four ways to buy AI voice-over: a per-character TTS API, a token-based newer speech model, a credit-based subscription workspace, and a per-minute/project finished voice-over service. Prices and units differ; a single “per million characters” number is not a full comparison.
| Service / plan | Billing unit | Official list price | Note |
|---|---|---|---|
| OpenAI tts-1 | Million characters | USD 15/million characters | Standard voice, input characters |
| OpenAI tts-1-hd | Million characters | USD 30/million characters | Higher audio quality |
| OpenAI gpt-4o-mini-tts | Token | USD 0.60/million text input tokens; USD 12/million audio output tokens | Not per character; not directly comparable |
| Google Standard / WaveNet | Million characters | USD 4/million characters | 0–4 million characters free |
| Google Neural2 / Polyglot | Million characters | USD 16/million characters | 0–1 million characters free |
| Google Chirp 3 HD | Million characters | USD 30/million characters | 0–1 million characters free |
| Google Studio | Million characters | USD 160/million characters | 0–1 million characters free |
| Google Instant custom voice | Million characters | USD 60/million characters | Instant custom voice |
| Google Gemini TTS | Token | USD 0.50–1/million text input tokens; USD 6–20/million audio output tokens | 25 tokens/second of audio; preview pricing through 31 Dec 2026 |
| ElevenLabs Starter | Subscription credits | USD 6/month, 30,000 credits | About 1 credit/character; USD 200/million |
| ElevenLabs Creator | Subscription credits | USD 22/month, 121,000 credits | About USD 181.82/million |
| ElevenLabs Pro | Subscription credits | USD 99/month, 600,000 credits | About USD 165/million |
| ElevenLabs Scale | Subscription credits | USD 299/month, 1.8M credits, 3 seats | About USD 166.11/million |
| ElevenLabs Business | Subscription credits | USD 990/month, 6M credits, 10 seats | About USD 165/million |
Official list prices in USD, excluding tax, as of 31 October 2026. OpenAI tts-1/tts-1-hd and Google legacy/Chirp/Studio prices are per character; OpenAI gpt-4o-mini-tts and Google Gemini TTS are token-based; ElevenLabs is a subscription credit model, estimated at 1 credit/character for the V2 multilingual model (actual models/discounts may differ). Azure TTS did not provide a stable public unit price and is excluded.
What do 100k, 1M, and 10M characters cost per month?
Per-character API: about USD 0.40–16 for 100k, USD 4–160 for 1M, and USD 40–1,600 for 10M. ElevenLabs starts at about USD 299 for 1M characters.
| Option | 100k characters | 1M characters | 10M characters | Billing |
|---|---|---|---|---|
| OpenAI tts-1 | USD 1.50 | USD 15 | USD 150 | Character |
| OpenAI tts-1-hd | USD 3 | USD 30 | USD 300 | Character |
| Google Standard/WaveNet | USD 0.40 | USD 4 | USD 40 | Character |
| Google Neural2 | USD 1.60 | USD 16 | USD 160 | Character |
| Google Chirp 3 HD | USD 3 | USD 30 | USD 300 | Character |
| Google Studio | USD 16 | USD 160 | USD 1,600 | Character |
| ElevenLabs Creator | USD 22 (includes 121k) | — | — | Subscription |
| ElevenLabs Scale | — | USD 299 (includes 1.8M) | — | Subscription |
| ElevenLabs Business | — | — | Custom (above 6M needs Enterprise) | Subscription |
Editorial model: characters × per-million price; excludes tax, commercial licensing, voice cloning, pronunciation review, audio storage, video production, and human listening. Google free tiers and ElevenLabs first-month discounts are not deducted. ElevenLabs is estimated at 1 credit/character; 1M characters needs Scale (1.8M credits); 10M exceeds Business’s 6M credits and needs Enterprise. “—” means the tier does not include enough credits.
What does a subscription credit cost per million characters?
ElevenLabs tiers are about USD 165–200 per million characters, 5–40× the per-character OpenAI/Google APIs; the premium pays for voice library, cloning, UI, and commercial licensing.
| ElevenLabs tier | Monthly fee | Included credits | USD per million characters | Fit |
|---|---|---|---|---|
| Starter | USD 6 | 30,000 | 200 | Small trial/individual |
| Creator | USD 22 | 121,000 | 181.82 | Short video/small team |
| Pro | USD 99 | 600,000 | 165 | Higher volume/pro quality |
| Scale | USD 299 | 1,800,000 | 166.11 | 3 seats/team collaboration |
| Business | USD 990 | 6,000,000 | 165 | 10 seats/low-latency API |
Estimated at 1 credit/character for the V2 multilingual model; V2 Flash/Turbo models may offer 0.5–1 credit/character, changing the effective price. Annual billing equals 10 months; excludes tax and overage top-ups. Subscriptions also include UI, voice library, cloning, team/commercial licensing, so a character-only comparison is incomplete.
Why is the API fee not the voice-over budget?
Commercial licensing, voice cloning/consent, pronunciation review, audio post, video production, and human listening are the main lines; TTS only turns text into speech.
| Cost line | Included in API/subscription? | How to meter | Common omission |
|---|---|---|---|
| TTS/subscription | Yes | Characters × price; credits/month | Mixing token and character pricing; top-ups |
| Commercial licence/voice rights | Partly | Scope, term, channel, territory | Using a celebrity/client voice without consent |
| Voice cloning/custom voice | Partly | Cloning runs, training/verification, seats | Reusing a clone across clients/projects |
| Pronunciation/multilingual review | No | Listening, phonetics/SSML, regeneration | Wrong model names, units, brand names |
| Audio/video post | No | Editing, mixing, subtitles, music | Counting TTS only, not post |
| Storage/CDN | No | Audio duration, bitrate, traffic | Keeping raw audio indefinitely |
| Human listening/QA | No | Audio minutes × listening speed × labour | Assuming generated audio can be published |
Editorial guidance, not an accounting taxonomy. Follow corporate finance policy, vendor contracts, and actual invoices.
How to choose by scenario
Standard narration: per-character API. Brand voice/cloning: subscription or custom voice. High volume: calculate per-million characters, licensing, and post first.
| Scenario | Start with | Why | Do not ignore |
|---|---|---|---|
| Product video/course standard narration | Per-character API (tts-1/Neural2) | Low unit price, batchable, scriptable | Pronunciation, pace, emotion, regeneration |
| Brand voice/multilingual | Chirp 3 HD/Studio or subscription library | Consistent voice, language coverage, control | Licence scope, cloning consent, channels |
| High-volume API integration | OpenAI/Google per character | Low marginal cost, easy CMS/video pipeline | Concurrency, retries, storage, character count |
| Need UI/team/cloning | ElevenLabs subscription | Voice library, cloning, seats, commercial licence | Credit overage, model discounts, annual billing |
| IVR/customer prompts | Low-latency TTS + pre-approval | Stable, updateable, multilingual | Number/address/amount reading, compliance |
| Celebrity/client voice clone | Consent first, then technology | Unconsented cloning is high risk | Written consent, use/term/withdrawal |
Editorial guidance, not a vendor ranking. Test 3–5 voices on the same script, language, and duration; compare pronunciation, naturalness, emotion, and regeneration cost.
What should procurement and acceptance measure?
Measure pronunciation (models/units/amounts), language/accent, pace/pauses, loudness, artefacts, commercial licence, voice consent, and cost per 1,000 characters — not a demo alone.
| Acceptance item | Suggested definition | Evidence |
|---|---|---|
| Pronunciation | Models, units, amounts, brand names, abbreviations | Script, phonetics/SSML, listening record |
| Language/accent | Target language/region/accent, native sampling | Voice list, sampling score |
| Pace/pauses/emotion | Matches video/course rhythm, adjustable | Timeline, parameters, version |
| Loudness/quality/artefacts | Consistent loudness, no clipping/hum/dropouts | Waveform, loudness report |
| Commercial licence | Channel, territory, term, derivative use | Contract/licence, invoice |
| Voice consent | Written consent and withdrawal for clones/real voices | Consent form, use record |
| Cost per 1,000 characters | (TTS/subscription + review + post) ÷ 1,000 characters | Invoice, hours, post record |
| Regeneration/versioning | Failed regeneration, script changes, traceability | Generation log, version record |
There is no universal naturalness score; product videos, IVR, courses, and audiobooks have different standards. High-risk/public content needs human listening, not only automated scoring.
Who should read this — and who should not?
For teams producing product videos, courses, IVR, or multilingual voice-over. Not for teams with no script/licence/listening process that only want the cheapest output.
Read if: you work in marketing, training, customer service, video/content, IT, or procurement; you are comparing OpenAI TTS, Google TTS, and ElevenLabs; you need to compare per-character, credit, and token pricing, cloning, and commercial licensing. Skip if: you have only one or two clips and no script/review; you want to clone a celebrity/client voice without consent; you treat TTS output as a finished video; you treat a subscription price as an API unit price. Supports: unit conversion, 100k/1M/10M-character models, subscription effective price, TCO, scenario selection, and acceptance metrics.
Lingyi does not resell voice APIs or claim any voice is “most natural”. If you need voice-over integrated with a video pipeline, CMS, private gateway, permissions, and audit, see LingHarness; for a few public product videos, a cloud vendor or subscription tool is usually faster. To scope a script, licence, and acceptance checklist, talk to us.
See our separate notes on speech-to-text cost, AI customer-service cost, and AI image commercial boundaries; this article covers TTS/voice-over character and subscription costs only.
FAQ
How do I compare a per-character API with ElevenLabs?
Divide the subscription fee by actual characters used to get an effective per-million price. ElevenLabs tiers are about USD 165–200/million characters; OpenAI/Google per-character APIs are about USD 4–30. The subscription includes UI, voice library, cloning, team seats, and commercial licensing. For pure API integration, per-character is usually cheaper; for a workspace/cloning/team, the subscription may be simpler.
Why is Google Studio 40× Standard?
Google lists Standard/WaveNet at USD 4/million characters and Studio at USD 160/million. Studio targets higher-quality/more natural voice-over; Standard suits prompts and batch narration. Test the same script before paying the premium.
Why can’t gpt-4o-mini-tts be compared per character?
It bills by token: USD 0.60/million text input tokens and USD 12/million audio output tokens, not characters. Character-to-token and audio-token conversion depends on language, tokenizer, and output duration; you cannot apply tts-1’s USD 15/million-character price. Estimate with the actual script and invoice.
What does 1M characters cost per month?
Per-character API: OpenAI tts-1 about USD 15, tts-1-hd 30, Google Standard 4, Neural2 16, Chirp 3 HD 30, Studio 160; ElevenLabs at 1M characters needs Scale at about USD 299 (1.8M credits). Add licensing, review, post, and listening; the table is an editorial model, not a quote.
What should we check before voice cloning?
Get written consent from the real person/client/celebrity, define use, channel, territory, term, and withdrawal; do not reuse a clone across clients/projects. Technical similarity is not legal permission; confirm the licence under contract and applicable law. This is not legal advice.
How should we accept voice-over quality?
Use the same script to test models, units, amounts, brand names, abbreviations, and target accent; check pace/pauses, loudness, artefacts, regeneration, and commercial licence. Product videos/IVR/courses have different standards; high-risk content needs human listening, not only vendor demos or automated scores.
Method and limitations
This article uses only official pricing/product pages accessible on 31 October 2026. It does not use private contracts, customer invoices, or vendor internal data. Prices, free tiers, models, credits, and commercial terms can change. The article has no actual enterprise agreement, tax, FX, cloning licence, review, or post-production data, so every monthly cost is an editorial model, not a quote. Azure TTS did not provide a stable public unit price and is excluded. OpenAI gpt-4o-mini-tts and Google Gemini TTS are token-based and not forced into a character price. ElevenLabs is estimated at 1 credit/character for the V2 multilingual model; actual models/discounts may differ.
Sources and basis
Data cut-off: 31 October 2026. The following are official pricing/product pages; prices, allowances, models, and terms follow the original page and contract.
- OpenAI, API Pricing: platform.openai.com/docs/pricing. tts-1/tts-1-hd per-million-character prices and gpt-4o-mini-tts token prices.
- Google Cloud, Text-to-Speech pricing: cloud.google.com/text-to-speech/pricing. Standard/WaveNet, Neural2, Chirp 3 HD, Studio, Instant custom voice, and Gemini TTS prices/free tiers.
- ElevenLabs, Pricing: elevenlabs.io/pricing. Starter/Creator/Pro/Scale/Business monthly fees, credits, 1 credit/character, commercial licensing/cloning.