AI KeyboardAndroid speech, offline SenseVoice and the on-device writing model cost nothing. This page only prices Groq and Gemini, where you pay the provider directly with your own key. Prices were checked in October 2026 and change; the provider's page always wins.
Models AI Keyboard uses
| Job | Model | Published price (paid) |
|---|---|---|
| Groq dictation | whisper-large-v3 | $0.111 per hour of audio |
| Groq writing styles | openai/gpt-oss-20b | $0.075 in / $0.30 out per 1M tokens |
| Gemini Live dictation | gemini-3.1-flash-live-preview | $3.00 per 1M audio input tokens, $4.50 per 1M text output tokens |
| Gemini writing styles | gemini-3.5-flash-lite | $0.30 in / $2.50 out per 1M tokens |
Estimator
Estimated monthly total
$—
The range reflects tokenisation uncertainty.
- Dictation
- —
- Writing styles
- —
- Transcript tokens
- —
- Gemini Live audio input tokens
- —
- Style input tokens
- —
- Style output tokens
- —
Quick examples
Paid-tier estimates with your speaking speed and style settings above, for different daily speech time.
| Per day | Android speech + Groq styles | Groq Whisper + Groq styles | Gemini Live + Gemini styles |
|---|---|---|---|
| Needs JavaScript. | |||
How this is calculated
- Gemini audio is counted at 32 tokens per second (Gemini token docs). Groq Whisper is billed per hour of audio.
- Transcript tokens come from words per minute, at roughly 100 tokens per 60–80 English words. Japanese and Chinese tokenise differently.
- The Gemini Live line assumes the transcript is billed as text output, which is the cautious choice.
- Writing styles send the transcript plus your prompt; the estimator ignores the prompt overhead and any minimum billing per request.
- Your real numbers are in Settings → Cloud & privacy, which lists tokens used request by request.