Every action in Speak AI draws on one credit balance, so the useful question is not “how many credits do I have” but “what does that buy”. This page gives you the rates and the arithmetic to answer that before you spend anything.
If you want the concepts first (what credits are, where to check your balance, what happens when you run low), read credits and usage.
What one credit buys
| What you do | One credit gets you |
|---|---|
| Transcribe uploaded audio or video | 1 hour |
| Transcribe with the Azure or AWS engines | 40 minutes |
| Record a meeting with the Meeting Assistant | 30 minutes |
| Translate a transcript | about 16,700 characters |
| Analyze text notes | 100 notes |
| Import a YouTube video | 20 imports |
| Ask AI Chat a question | about 10 questions |
Transcription bills by the minute and is never rounded up to the hour, so a 12-minute file costs 0.2 credits, not a whole one.
What a balance covers
Pick the row closest to your balance. Every column assumes you spend the whole balance on that one thing, so treat these as ceilings rather than a budget you can hit all at once.
| Credits | Transcription | Meeting Assistant | Translation | AI Chat questions |
|---|---|---|---|---|
| 2 (the free trial) | 2 hours | 1 hour | 33,000 characters | 20 |
| 10 | 10 hours | 5 hours | 167,000 characters | 100 |
| 25 | 25 hours | 12 hours 30 minutes | 417,000 characters | 250 |
| 100 | 100 hours | 50 hours | 1,670,000 characters | 1,000 |
The AI Chat column uses the cautious planning rate of 0.1 credits a question, which is the same figure your Usage screen shows. What a question actually costs depends on the model.
AI Chat: standard versus premium models
Speak charges a chat question on what the model itself costs, so the gap between a standard model and a premium one is large. These are averages measured over 30 days of real Speak traffic, not list prices.
| Model | Tier | Credits a question | Questions per 10 credits |
|---|---|---|---|
| Gemini 2.5 Flash | Standard | about 0.04 | about 250 |
| GPT-5.5 | Premium | about 0.63 | about 16 |
| Claude Sonnet | Premium | about 0.75 | about 13 |
| Claude Opus | Premium | about 1.20 | about 8 |
A long question that sends a large transcript to the model costs more than a short one, and a short one costs less, so read these as typical rather than fixed.
Work out what a specific job costs
Three formulas cover almost everything:
- Transcription: minutes of audio ÷ 60. Use ÷ 40 on the Azure or AWS engines.
- Meeting Assistant: minutes of meeting ÷ 30.
- Translation: characters ÷ 16,700. If you only know the length of the recording, a minute of speech is roughly 800 characters.
A worked example. You have a 45-minute interview and you want it transcribed, translated, and then a handful of questions asked about it on the standard model:
| Step | Arithmetic | Credits |
|---|---|---|
| Transcribe 45 minutes | 45 ÷ 60 | 0.75 |
| Translate the transcript | 45 × 800 = 36,000 characters, ÷ 16,700 | 2.16 |
| Ask 10 questions on Gemini 2.5 Flash | 10 × 0.04 | 0.40 |
| Total | about 3.3 |
Run the same interview through Claude Opus instead of Gemini and those 10 questions cost about 12 credits on their own, which is four times the rest of the job put together.
Pay-As-You-Go doubles every rate
On the Pay-As-You-Go plan every credit rate on this page doubles, so a credit buys half as much: 30 minutes of transcription instead of an hour, 15 minutes of Meeting Assistant instead of 30. Halve every figure in the tables above and you have your answer.
What does not draw on credits
Reviewing transcripts, reading insights, sharing, and exporting are free. Only processing new work costs credits, so you can go back through everything you have already made as often as you like.
Re-transcribing a file is new work: it runs the whole pipeline again and charges again.
For current credit amounts and per-seat pricing, see speakai.co/pricing.
Related: Credits and usage · Free trial · Plans and pricing · Payment methods