What a session costs: subscription, credits and how they are spent
GPT Realtime Assistant is a subscription with AI credits included, and credits are spent on what a session actually uses. That is fairer than a flat per-minute price, and harder to predict. Here is how it works, and what makes a session cheaper.
The subscription
The app needs an active monthly subscription to run: $9.99 a month in the US. Every template, translation and readback are included. Each month comes with 2,000 AI credits.
A thousand credits are one US dollar of model usage at OpenAI's list rates, metered by the audio and text the session actually uses.
When credits run low
Credit packs add more without changing the subscription:
- 1,000 credits for $4.99 — about one US dollar of model usage.
- 5,000 credits for $19.99 — about five US dollars of model usage.
- Pack credits never expire while the subscription is active.
What a session spends
Credits follow what the model does. Audio in and out costs more than text. Context the model has already seen, and can reuse, costs less. A session reserves a small amount when it starts and refunds what it did not use when it ends.
Making credits last
Because audio costs more than text, the biggest levers are the ones that change how much audio the model handles:
- Use brief answers during live conversations. Shorter answers are less output.
- End the session when the conversation ends, rather than leaving it listening to an empty room.
- On iPhone and iPad, transcription can come from Apple's on-device speech recognition instead of the model.
Managing and cancelling
Purchases go through Apple. Manage or cancel the subscription in Settings, under your name, then Subscriptions. Restore purchases works on any of your devices. Refunds are handled by Apple, at reportaproblem.apple.com.
How many credits come with the subscription?
2,000 each month. A thousand credits are one US dollar of model usage at OpenAI's list rates.
Do unused credits expire?
Pack credits never expire while the subscription is active.
What costs the most credits?
Audio, in and out. Text costs less, and reused context costs less still.
