Speech and audio API pricing
Text to speech, speech to text and music generation: 33 models from 7 companies. The cheapest voice is Gemini 3.8 Flash-Lite TTS at $0.009 a minute; transcription starts at $0.0025 a minute (Universal-2). Each price is checked on the company’s own pricing page, 4 Nov 2026.
Jump to: Text to speech · Speech to text · Music
Text to speech API prices
Cheapest first
| Model | Billed | Price | An hour of speech | Source |
|---|---|---|---|---|
 Gemini 3.8 Flash-Lite TTSGoogle · gemini-3.8-flash-lite-tts | Per minute of audio | $0.009company’s estimate · $0.018 from 1 Jan 2027 | $0.54 | Source for Gemini 3.8 Flash-Lite TTS: Gemini API pricing (opens in a new tab) |
 Voxtral TTS 3Mistral AI | Per 1,000 characters | $0.013 | $0.70estimate | Source for Voxtral TTS 3: Mistral API pricing (opens in a new tab) |
Gemini 3.8 Flash TTSGoogle · gemini-3.8-flash-tts | Per minute of audio | $0.0135company’s estimate · $0.027 from 1 Jan 2027 | $0.81 | Source for Gemini 3.8 Flash TTS: Gemini API pricing (opens in a new tab) |
| AssemblyAI Text-to-SpeechAssemblyAI | Per 1,000 characters | $0.015 | $0.81estimate | Source for AssemblyAI Text-to-Speech: AssemblyAI pricing (opens in a new tab) |
| Aura-1Deepgram | Per 1,000 characters | $0.015 | $0.81estimate | Source for Aura-1: Deepgram pricing (opens in a new tab) |
| Aura-2Deepgram | Per 1,000 characters | $0.03 | $1.62estimate | Source for Aura-2: Deepgram pricing (opens in a new tab) |
Gemini 3.1 Flash TTS (preview)Google · gemini-3.1-flash-tts-preview | Per minute of audio | $0.03from its token rate | $1.80 | Source for Gemini 3.1 Flash TTS (preview): Gemini API pricing (opens in a new tab) |
| Eleven Flash / TurboElevenLabs | Per 1,000 characters | $0.04 | $2.16estimate | Source for Eleven Flash / Turbo: ElevenLabs API pricing (opens in a new tab) |
| Eleven v4 TurboElevenLabs | Per 1,000 characters | $0.04 | $2.16estimate | Source for Eleven v4 Turbo: ElevenLabs API pricing (opens in a new tab) |
| Flux TTSDeepgram | Per 1,000 characters | $0.045 | $2.43estimate | Source for Flux TTS: Deepgram pricing (opens in a new tab) |
 Speech 2.8 TurboMiniMax · speech-2.8-turbo | Per 1,000 characters | $0.06 | $3.24estimate | Source for Speech 2.8 Turbo: MiniMax pricing (opens in a new tab) |
| Eleven v3ElevenLabs | Per 1,000 characters | $0.08 | $4.32estimate | Source for Eleven v3: ElevenLabs API pricing (opens in a new tab) |
| Eleven v4ElevenLabs | Per 1,000 characters | $0.08 | $4.32estimate | Source for Eleven v4: ElevenLabs API pricing (opens in a new tab) |
Speech 2.8 HDMiniMax · speech-2.8-hd | Per 1,000 characters | $0.10 | $5.40estimate | Source for Speech 2.8 HD: MiniMax pricing (opens in a new tab) |
US dollar pay-as-you-go prices before tax, checked 4 Nov 2026. An hour of speech is 60 minutes of audio; for per-character prices it assumes about 900 characters a minute, a typical speaking rate, so it’s an estimate. Gemini voices are billed per token and Google publishes the per-second equivalent we use.
Speech to text (transcription) API prices
Cheapest first
| Model | Audio | Per minute | Per hour | Source |
|---|---|---|---|---|
| Universal-2AssemblyAI | Recorded files | $0.0025 | $0.15listed as $0.15 an hour | Source for Universal-2: AssemblyAI pricing (opens in a new tab) |
 GPT-4o mini TranscribeOpenAI · gpt-4o-mini-transcribe | Recorded files | $0.003company’s estimate | $0.18 | Source for GPT-4o mini Transcribe: OpenAI API pricing (opens in a new tab) |
| Universal-3.5 ProAssemblyAI | Recorded files | $0.0035 | $0.21listed as $0.21 an hour | Source for Universal-3.5 Pro: AssemblyAI pricing (opens in a new tab) |
| Voxtral Mini Transcribe 3Mistral AI | Recorded files | $0.0035 | $0.21 | Source for Voxtral Mini Transcribe 3: Mistral API pricing (opens in a new tab) |
| Scribe v2ElevenLabs | Recorded files | $0.00367 | $0.22listed as $0.22 an hour | Source for Scribe v2: ElevenLabs API pricing (opens in a new tab) |
| Voxtral Small Transcribe 3Mistral AI | Recorded files | $0.004 | $0.24 | Source for Voxtral Small Transcribe 3: Mistral API pricing (opens in a new tab) |
GPT TranscribeOpenAI · gpt-transcribe | Recorded files | $0.0045company’s estimate | $0.27 | Source for GPT Transcribe: OpenAI API pricing (opens in a new tab) |
Gemini 3.5 TranscribeGoogle · gemini-3.5-transcribe | Recorded files | $0.005company’s estimate | $0.30 | Source for Gemini 3.5 Transcribe: Gemini API pricing (opens in a new tab) |
GPT-4o TranscribeOpenAI · gpt-4o-transcribe | Recorded files | $0.006company’s estimate | $0.36 | Source for GPT-4o Transcribe: OpenAI API pricing (opens in a new tab) |
| MiniMax Speech-to-TextMiniMax | Recorded files | $0.00633 | $0.38listed as $0.38 an hour | Source for MiniMax Speech-to-Text: MiniMax pricing (opens in a new tab) |
| Scribe v2 RealtimeElevenLabs | Live stream | $0.0065 | $0.39listed as $0.39 an hour | Source for Scribe v2 Realtime: ElevenLabs API pricing (opens in a new tab) |
| Flux STTDeepgram | Live stream | $0.0077 | $0.46 | Source for Flux STT: Deepgram pricing (opens in a new tab) |
| Nova-3Deepgram | Live stream | $0.0077 | $0.46 | Source for Nova-3: Deepgram pricing (opens in a new tab) |
| Voxtral Mini Transcribe Realtime 3Mistral AI | Live stream | $0.008 | $0.48 | Source for Voxtral Mini Transcribe Realtime 3: Mistral API pricing (opens in a new tab) |
Gemini 3.5 Transcribe LiveGoogle · gemini-3.5-transcribe-live | Live stream | $0.009company’s estimate | $0.54 | Source for Gemini 3.5 Transcribe Live: Gemini API pricing (opens in a new tab) |
GPT Live TranscribeOpenAI · gpt-live-transcribe | Live stream | $0.017company’s estimate | $1.02 | Source for GPT Live Transcribe: OpenAI API pricing (opens in a new tab) |
Per minute of audio processed. Where a company prices by the hour, the per-minute figure is that price divided by 60. OpenAI and Google bill transcription per token and publish the per-minute estimates shown.
Music generation API prices
| Model | Price | Per | Source |
|---|---|---|---|
Lyria 3 Clip (preview)Google · lyria-3-clip-preview | $0.04 | 30-second clip | Source for Lyria 3 Clip (preview): Gemini API pricing (opens in a new tab) |
Lyria 3.5Google · lyria-3.5 | $0.08 | full song | Source for Lyria 3.5: Gemini API pricing (opens in a new tab) |
| Eleven MusicElevenLabs | $0.15 | minute | Source for Eleven Music: ElevenLabs API pricing (opens in a new tab) |
What each price includes
- Gemini 3.8 Flash-Lite TTS Google
- Billed per token: $6 per 1M audio output tokens until 31 December 2026, then $12. Google states this equals $0.0015 per 10 seconds of audio ($0.003 from 1 January). Text input costs $0.50 per 1M tokens.
- Voxtral TTS 3 Mistral AI
- $13 per 1M characters; cached input $1.30 per 1M.
- Gemini 3.8 Flash TTS Google
- Billed per token: $9 per 1M audio output tokens until 31 December 2026, then $18. Google states this equals $0.00225 per 10 seconds of audio ($0.0045 from 1 January). Text input costs $0.50 per 1M tokens.
- AssemblyAI Text-to-Speech AssemblyAI
- $15 per 1M input characters. Cloned voices and EU processing cost the same.
- Aura-1 Deepgram
- Pay-as-you-go rate; $0.0135 per 1,000 characters on the Growth plan.
- Aura-2 Deepgram
- Pay-as-you-go rate; $0.027 per 1,000 characters on the Growth plan.
- Gemini 3.1 Flash TTS (preview) Google
- Billed per token: $20 per 1M audio output tokens, and Google counts 25 tokens per second of audio, which works out at $0.03 a minute. Text input costs $1 per 1M tokens.
- Eleven Flash / Turbo ElevenLabs
- ElevenLabs’ fast, low-cost models (about 75 ms latency), up to 40,000 characters a request.
- Eleven v4 Turbo ElevenLabs
- v4 expressiveness tuned for real-time use and voice agents.
- Flux TTS Deepgram
- Same rate on pay-as-you-go and Growth.
- Speech 2.8 Turbo MiniMax
- $60 per 1M characters, synchronous or asynchronous (up to 1M characters a request).
- Eleven v3 ElevenLabs
- Expressive delivery with multi-speaker dialogue, up to 5,000 characters a request. Eleven v2 Multilingual costs the same.
- Eleven v4 ElevenLabs
- ElevenLabs’ most expressive model, with audio tags.
- Speech 2.8 HD MiniMax
- $100 per 1M characters. Designed voices cost $3 each and cloned voices $1.50 each, charged on first use.
- Universal-2 AssemblyAI
- Pre-recorded audio in 99 languages.
- GPT-4o mini Transcribe OpenAI
- Billed per token ($1.25 per 1M audio input tokens, $5 per 1M text output); OpenAI estimates $0.003 a minute.
- Universal-3.5 Pro AssemblyAI
- AssemblyAI’s most accurate pre-recorded model, 18 languages, with speaker diarization.
- Voxtral Mini Transcribe 3 Mistral AI
- Cached input $0.00035 a minute.
- Scribe v2 ElevenLabs
- 90+ languages, keyterm prompting. Scribe v2 Medical costs the same.
- Voxtral Small Transcribe 3 Mistral AI
- Cached input $0.0004 a minute.
- GPT Transcribe OpenAI
- OpenAI lists an estimated $0.0045 a minute.
- Gemini 3.5 Transcribe Google
- Billed per token ($2 per 1M audio input tokens, $12 per 1M text output); Google estimates about $0.005 a minute. Includes language detection, speaker diarization and word timestamps.
- GPT-4o Transcribe OpenAI
- Billed per token ($2.50 per 1M audio input tokens, $10 per 1M text output); OpenAI estimates $0.006 a minute.
- MiniMax Speech-to-Text MiniMax
- With streaming, speaker diarization and subtitle export.
- Scribe v2 Realtime ElevenLabs
- About 150 ms latency.
- Flux STT Deepgram
- Streaming, English, pay-as-you-go ($0.0065 on Growth); multilingual $0.0078.
- Nova-3 Deepgram
- Streaming, one language, pay-as-you-go ($0.0065 on Growth); multilingual $0.0092. Deepgram says pre-recorded audio costs less. Billed per second.
- Voxtral Mini Transcribe Realtime 3 Mistral AI
- Cached input $0.0008 a minute.
- Gemini 3.5 Transcribe Live Google
- Real-time over WebSockets, billed per token ($3.50 per 1M audio input tokens, $21 per 1M text output); Google estimates about $0.009 a minute.
- GPT Live Transcribe OpenAI
- OpenAI lists an estimated $0.017 a minute; gpt-realtime-whisper costs the same.
- Lyria 3.5 Google
- Lyria 3 Pro (preview) costs the same.
- Eleven Music ElevenLabs
- Up to 5 minutes a track; commercial use licensing on Starter plans and above.
How speech APIs charge
- Voices by the character or the minute. Most text-to-speech APIs charge per 1,000 characters of text you send. Google charges per audio token and publishes what a second of audio costs, so its prices are per minute.
- Transcription by the minute. Speech to text is billed per minute (or hour) of audio. Live streaming usually costs more than sending a recorded file.
- Extras can cost extra. Speaker labels, redaction and keyword boosting are separate charges at some companies (Deepgram lists each one) and included at others, such as Gemini 3.5 Transcribe.
- Plans and free credit. Deepgram’s prepaid Growth plan and ElevenLabs’ subscriptions cost less per unit than pay-as-you-go. Deepgram gives $200 of free credit, and AssemblyAI’s free tier covers up to 185 hours of recorded audio.
Questions
What is the cheapest text-to-speech API?
Gemini 3.8 Flash-Lite TTS from Google, at $0.009 a minute of audio, rising to $0.018 on 1 Jan 2027, about $0.54 an hour. Among APIs priced per character, the cheapest is Voxtral TTS 3 at $0.013 per 1,000 characters.
What is the cheapest speech-to-text API?
For recorded files, Universal-2 from AssemblyAI at $0.0025 a minute ($0.15 an hour). For live transcription, Scribe v2 Realtime at $0.0065 a minute. Accuracy, languages and features such as speaker labels differ, so test on your own audio.
How much does the ElevenLabs API cost?
Pay-as-you-go text to speech costs $0.04 and $0.08 per 1,000 characters: Eleven Flash / Turbo $0.04, Eleven v4 Turbo $0.04, Eleven v3 $0.08 and Eleven v4 $0.08. Scribe v2 speech to text costs $0.22 an hour and Eleven Music $0.15 a minute. Subscription plans include characters each month.
How do I compare per-character and per-minute prices?
Speech runs at roughly 900 characters a minute, so 1,000 characters is a little over a minute of audio. On that basis $0.04 per 1,000 characters comes to about $2.16 an hour. Fast or slow voices change this, which is why we mark those hour figures as estimates.
Will Gemini text to speech get more expensive?
Yes. Gemini 3.8 Flash-Lite TTS goes from $0.009 to $0.018 a minute and Gemini 3.8 Flash TTS goes from $0.0135 to $0.027 a minute on 1 January 2027, when Google doubles its audio output rate.
What about OpenAI text to speech?
OpenAI has deprecated tts-1, tts-1-hd and gpt-4o-mini-tts, which shut down on 6 January 2027, and no longer lists their prices. It points developers to its realtime models instead, which are billed per audio token rather than per minute.
When were these prices checked?
On 4 Nov 2026, on each company’s own pricing page, linked from every row. Prices are US dollar pay-as-you-go rates before tax; volume plans cost less.
More pricing
Sources: Gemini API pricing, Mistral API pricing, AssemblyAI pricing, Deepgram pricing, ElevenLabs API pricing, MiniMax pricing, OpenAI API pricing. Checked 4 Nov 2026.