Cartesia's Free plan includes 20,000 credits a month and Google's Gemini 3.8 Flash TTS charges $9.00 per million audio tokens, and neither number means anything until it is turned into seconds of speech. Text to speech vendors bill by the character, the credit, the token, the minute, or a plan allowance, and each unit hides a different question.
Bland Speech bills by the character, $0.04 per 1,000 ($40 per million), pay as you go with no monthly plan. That is the simplest model to forecast, though not the lowest-priced.
What are the main ways text to speech is priced?
As of November 2026, text to speech is priced six ways: per character, per UTF-8 byte, per credit with a monthly allowance, per audio token, per minute, or against a prepaid commitment, and a rate in one unit can't be compared with a rate in another until both become dollars per hour of audio.
What these rows mean
- Per character: the provider multiplies the characters of text sent by a rate. Bland Speech charges the same $0.04 per 1,000 in its studio and API, directed or not.
- Per UTF-8 byte: Fish Audio says 1M UTF-8 bytes is about 180,000 English words, or about 12 hours of speech.
- Credits: Cartesia, Gradium, and Typecast each charge one credit per character, so a credit equals a character at all three, and the plan sets what a credit costs.
- Per audio token: Google says audio tokens correspond to 25 tokens per second of audio.
- Per minute: Rime bills Mist v3 at $0.030 a minute, Coda v2 at $0.050, and Prism at $0.080 on its Flex plan.
- Prepaid commitment: Azure's commitment tiers exclude HD, Azure OpenAI, custom, and personal voices.
How does per-character pricing work, and what counts as a character?
Per-character pricing multiplies the characters of text sent by a rate, and providers differ on whether pauses, markup, spaces, and non-English letters count.
Amazon Polly charges for the characters of text converted to speech. Speechify bills the characters actually spoken and doesn't count SSML tags (SSML is a markup language for controlling pauses and pronunciation). ReadSpeaker's speechCloud counts one credit per character excluding whitespace. Cartesia counts pauses too: each break tag counts as one credit, and adding an accent to a voice costs 225 credits. Fish Audio counts UTF-8 bytes instead of characters, so a language whose letters take two or three bytes costs two or three times as much per letter as English text.
Bland Speech bills each character against a wallet balance, and credits don't expire. Each request carries a minimum charge of $0.001, so very short generations bill at the minimum, and some public-library voices add a per-character creator fee. A fully delivered request is charged for its whole text, and a synthesis that fails before any audio is written isn't charged. Per-character billing has one real strength for forecasting, because a script of 4,000 characters costs $0.16 on Bland Speech whatever the month before looked like.
How do credit prices work, and what does a credit buy?
A credit is whatever the provider defines it as, and at Cartesia, Gradium, and Typecast it equals one character, with a monthly plan fee deciding what each credit costs.
Cartesia sells Free at 20,000 credits a month, Pro at $5 for 100,000, Startup at $49 for 1.25 million, and Scale at $299 for 8 million, with overage rates of $65, $45, and $38 per million credits on the paid plans. Cartesia's own conversion is about 750 to 800 credits a minute, so its Free plan holds about 27 minutes. Overages are opt-in and billed when a plan renews, and credits roll over up to twice the monthly amount.
Credit plans cross a flat per-character rate at a predictable volume. A month of 100,000 characters costs $4 on Bland Speech and $5 on Cartesia Pro. A month of 2 million characters costs $80 on Bland Speech and $82.75 on Cartesia Startup ($49 plus 750,000 overage credits at $45 per million). A month of 20 million characters costs $800 on Bland Speech and $755 on Cartesia Scale ($299 plus 12 million overage credits at $38 per million). A plan beats the flat rate only when a team fills a large allowance: Cartesia Pro's $5 for 100,000 credits works out to $50 per million even used in full, while Scale's $299 for 8 million is about $37 per million.
Several providers return a plan fee as credit. Inworld's Creator plan is $25 a month with $25 in credits and lowers TTS-2 from $25 to $20 per million characters, with credits rolling over for three months. Gradium's plan credits must be used in the current month and don't roll over, while credits bought pay as you go do. Unreal Speech's Starter plan is $10 a month for 500,000 characters, with overage at $20 per million and unused characters rolling over on paid plans. Typecast bills overage in blocks, $9 per 100,000 credits on Lite.
The same plan can show two different counts. ElevenLabs shows Creator as $22 a month for 121,000 credits on its main pricing page and as $22 a month for 275,000 characters of Eleven v4 on its API pricing page, so a buyer has to read which page and which model. ElevenLabs' API rates are in dollars per character, $0.08 per 1,000 for Eleven v4 and $0.04 for v4 Turbo.
Free tiers make the unit gap visible, and each figure here is the company's own: Cartesia's Free plan is about 27 minutes a month, ElevenLabs' Free plan is 10,000 characters of Eleven v4 and labelled about 10 minutes, Gradium's Free plan is 45,000 credits a month and about an hour, Inworld's On-Demand plan includes up to 70 minutes of TTS, and Bland Speech's 50,000 free characters are nearly an hour, once per new account.
How do token prices work, and how many seconds is a token?
A token is a chunk of data that a model reads or writes, and audio token pricing bills the audio that comes out, at a rate per million tokens, with text input billed on top.
Google is the clearest about the conversion: its audio tokens correspond to 25 tokens per second of audio. An hour is 3,600 seconds, which is 90,000 tokens, and at Gemini 2.5 Flash TTS's $10.00 per million audio tokens that is $0.90 for an hour of audio, plus text input at $0.50 per million tokens. Gemini 2.5 Pro TTS and Gemini 3.1 Flash TTS charge $20.00 per million audio tokens, which makes an hour $1.80. Gemini 3.8 Flash TTS costs $9.00 per million through 31 December 2026 and $18.00 from 1 January 2027, a price that doubles on a date.
Google bills Chirp 3: HD, a different voice family, by the character at $30 per million, so one provider needs two sets of arithmetic. Alibaba does the same: Qwen-Audio-3.1-TTS-Plus is $0.3 per 10,000 characters in Singapore, and Qwen-Audio-3.1-TTS-Flash is $1.87 per million output tokens plus $0.23 per million input tokens.
OpenAI prices gpt-4o-mini-tts at $0.60 per million text input tokens and $12.00 per million audio output tokens, and OpenAI will shut down gpt-4o-mini-tts, tts-1, and tts-1-hd on 6 January 2027. OpenAI's replacement path, gpt-realtime-2.1-mini, is priced at $10.00 per million audio input tokens and $20.00 per million audio output tokens over the Realtime API.
Which pricing model suits a phone line?
For a phone line, per-character pricing with no monthly plan suits best, because call volume swings from week to week and a plan allowance is either wasted or exceeded.
Bland Speech is built that way, at $0.04 per 1,000 characters billed against a prepaid balance that doesn't expire. A team can forecast a month by counting the words in its scripts, with no allowance to outgrow and no tokens per second to look up. The trade is volume discounts: Azure's commitment tier of $15,000 for 2,000 million characters works out to $7.50 per million against $15 pay as you go, Inworld's Growth plan cuts TTS-2 to $12.50 per million, and Deepgram's Growth plan takes 10% off Aura-2. A team already committed to tens of millions of characters a month can compare those plans with Bland Speech's single rate on its own volume.
Bland Speech v3, the voice behind that rate, ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, and the text to speech API sells it at that one price. The tables in text to speech API pricing and what an hour of AI speech costs turn these units into monthly and hourly bills, and Bland Speech has the studio and the rate. The wider set sits in Bland's text to speech comparisons and guides.
Frequently asked questions
What is the difference between per-character and per-token text to speech pricing?
Per-character pricing bills the text a team sends, while per-token pricing bills the audio a model produces. A character count is known before a request is sent, and an audio token count depends on how long the speech runs, at 25 tokens a second for Google's Gemini voices.
How much is a text to speech credit worth?
A credit is worth whatever a plan sets. At Cartesia, Gradium, and Typecast a credit equals one character, and Cartesia's overage runs from $65 per million credits on Pro to $38 on Scale, while Typecast's Lite plan prices a credit at $0.075 per 1,000. Bland Speech has no credits to convert, because its rate is $0.04 per 1,000 characters.
Do unused text to speech credits roll over?
Each provider decides. Cartesia lets a team hold up to twice its monthly credits, Inworld rolls credits over for three months, Unreal Speech rolls unused characters over on paid plans, and Gradium's plan credits expire at the end of the month. Bland Speech's credits don't expire.
Does a pause cost extra in text to speech?
At Cartesia, yes: each break tag counts as one credit. At Speechify, SSML tags don't count toward the bill, because Speechify bills the characters spoken.
What does Bland Speech charge for a very short request?
Each Bland Speech request carries a minimum charge of $0.001, so very short generations bill at the minimum instead of the per-character rate. A synthesis that fails before any audio is written isn't charged.
Is Bland Speech pay as you go?
Yes. Bland Speech bills $0.04 per 1,000 characters against a wallet balance, with no monthly plan, and a new account gets 50,000 characters free once.
