
Speech API Pricing [Nov 2026]
Scope: text-to-speech (TTS) and transcription (STT) rates, normalised to one minute and one hour. Vendors bill in four different units (per 1M characters, per 1M audio tokens, per minute, per hour), so the raw numbers are not comparable. Conversion assumptions are in the collapsed block below (official pages fetched 2026-11-15).
Bottom line (3 lines)
- Japanese narration → MAI-Voice-2.1-Flash ($0.0045/min). OpenAI tts-1 costs the same but is deprecated, so it is not the current pick.
- Transcription of recordings → MAI-Transcribe-2 ($0.10/hour), a promotional rate valid until the end of this year. Next is Meta Muse Voice Transcribe at $0.18/hour.
- Realtime conversation → an order of magnitude more expensive per minute (from $0.023/min). Do not compare it in the same table as file transcription.
Why normalisation is required
Per 1M charactersOpenAI tts-1
MAI-Voice
MAI-Voice
+
Per 1M audio tokensGemini TTS
(25 tokens/sec)
(25 tokens/sec)
+
Per hourTranscribe family
→
Common floorTTS: per min / 1,000 chars
STT: per hour
STT: per hour
Comparing "$9 vs $15" without normalising makes the cheaper model look three times more expensive.
Text to speech (TTS)
| Model | Per minute | Status |
|---|---|---|
| MAI-Voice-2.1-Flash | $0.0045 | Current |
| OpenAI tts-1 | $0.0045 | Deprecated |
| MAI-Voice-2.1 | $0.0066 | Current |
| OpenAI tts-1-hd | $0.0090 | Deprecated |
| Gemini 3.8 Flash-Lite TTS | $0.0090 → $0.0180 | Change 2027-01-01 |
| Gemini 3.8 Flash TTS | $0.0135 → $0.0270 | Change 2027-01-01 |
| OpenAI gpt-4o-mini-tts | ~$0.018 | Deprecated, estimate |
| Gemini 3.1 Flash TTS Preview | $0.0300 | Preview |
The same numbers as bars (per minute)
Per-1,000-character rates and the raw billing units (breakdown)
TTS: per 1,000 characters (1,000 characters = 3 min 20 s)
| Model | Per 1,000 chars |
|---|---|
| MAI-Voice-2.1-Flash | $0.0150 |
| OpenAI tts-1 | $0.0150 |
| MAI-Voice-2.1 | $0.0220 |
| OpenAI tts-1-hd | $0.0300 |
| Gemini 3.8 Flash-Lite TTS | $0.0300 → $0.0600 |
| Gemini 3.8 Flash TTS | $0.0450 → $0.0900 |
| OpenAI gpt-4o-mini-tts | ~$0.060 |
| Gemini 3.1 Flash TTS Preview | $0.1000 |
STT: raw billing unit (the table above is already normalised to an hour)
| Model | Billing unit |
|---|---|
| MAI-Transcribe-2 | Hour |
| Meta Muse Voice Transcribe | Hour (per second, floored) |
| Gemini 3.5 Transcribe | Tokens |
| Gemini 3.5 Transcribe Live | Tokens |
| MAI-Transcribe-2-Streaming | Hour |
| OpenAI gpt-live-transcribe | Minute ($0.017/min) |
Transcription (STT)
| Model | Per hour | Status |
|---|---|---|
| MAI-Transcribe-2 | $0.10 | Promo until year end |
| Meta Muse Voice Transcribe | $0.18 | Current |
| Gemini 3.5 Transcribe | $0.18 (est. $0.30) | Current |
| Gemini 3.5 Transcribe Live | $0.30 (est. $0.54) | Current (bidirectional) |
| MAI-Transcribe-2-Streaming | $0.54 | Introductory price |
| OpenAI gpt-live-transcribe | $1.02 | Current |
Caveats (4)
- OpenAI's three character-billed TTS models (tts-1 / tts-1-hd / gpt-4o-mini-tts) are deprecated. They stay in the table for comparison, not as a recommendation.
- "300 characters per minute" is this site's assumption (slow Japanese narration). English is roughly three times faster, so character-billed models consume more characters per minute.
- Two Gemini TTS models double in price on 2027-01-01 (the right-hand figure in the table).
- OpenAI does not publish a seconds-to-audio-token conversion for gpt-4o-mini-tts, so its per-minute figure is an estimate.
Conversion assumptions and the official wording (click to open)
- Audio tokens: 25 tokens per second of audio (1,500 per minute) — Google's note
- Characters to duration: 300 characters per minute (5 per second) — this site's assumption; vendors do not publish a speaking rate
- 1,000 characters = 3 min 20 s (200 s); 1 hour = 18,000 characters — derived above
- USD only, no FX conversion
Audio tokens correspond to 25 tokens per second of audio.
— ai.google.dev/gemini-api/docs/pricing (fetched 2026-11-15)