Core Workflow Rule 1,000 credits equal roughly $1 USD of OpenAI Realtime model usage; your $9.99/month plan includes 2,000 credits, unused session reservations refund automatically when you tap End, and local Whisper + Qwen sessions on Mac spend zero credits.
GPT Realtime Assistant Plans & Credit Mechanics (US App Store)
Plan / OptionUS App Store PriceCredits IncludedKey Rule
Monthly Subscription$9.99 / month2,000 AI credits / monthRequired to run the app across iPhone, iPad, and Mac
1,000-Credit Pack$4.99 one-time+1,000 consumable creditsTops up balance anytime without changing monthly plan
5,000-Credit Pack$19.99 one-time+5,000 consumable creditsBest per-credit top-up for heavy live voice/translation weeks
Local Mac Mode (Whisper + Qwen)Included with subscription0 credits spentRuns 100% on Apple Silicon Mac with text answers

Why Audio Minutes Cost More Than Text Tokens

Streaming live microphone audio into OpenAI's Realtime API and synthesizing expressive low-latency speech back out processes continuous acoustic tokens every second. Two-way simultaneous voice translation runs two live translation pipelines at once and uses roughly 85 credits per minute—meaning the 2,000 monthly credits cover around 23 minutes of non-stop bidirectional spoken interpretation.

By contrast, sessions where the assistant listens via Apple's on-device speech recognition or replies with concise on-screen text spend dramatically fewer credits per minute.

How Start-of-Session Reservation and Refunds Work

When you press Start, the backend reserves a small credit buffer to authorize your ephemeral OpenAI session token. Throughout the conversation, your live credit counter stays visible right on screen—never buried three menus deep.

The moment you press End, exact token consumption is settled and every unused reserved credit is refunded immediately to your balance.

Four Practical Ways to Stretch 2,000 Monthly Credits

Small workflow habits multiply how many meetings and study sessions fit inside your monthly allowance:

  1. Switch answer length to Brief during fast calls so the model doesn't generate long spoken monologues.
  2. Turn off spoken audio replies when reading side-by-side text answers on screen is sufficient (such as live lecture or meeting notes).
  3. Press End promptly when a call wraps up rather than leaving an idle microphone streaming room noise.
  4. On an Apple Silicon Mac, switch routine note-taking and private brainstorms to downloaded local Whisper + Qwen models, which use zero credits.

Pre-Session Checklist

Saved locally in your browser so you can check your setup before starting.

Frequently Asked Questions

Is GPT Realtime Assistant affiliated with OpenAI?

No. GPT Realtime Assistant is built independently by Arte Soft EOOD and uses OpenAI's Realtime API as its cloud model provider.

What happens if I run low on credits before my monthly renewal?

You can add a 1,000-credit ($4.99 US) or 5,000-credit ($19.99 US) consumable pack through Apple In-App Purchase at any time without altering your monthly renewal date.

Continue Exploring