| Dimension | Cloud Mode (OpenAI Realtime API) | Local Mac Mode (Whisper + Qwen) |
|---|---|---|
| Supported hardware | iPhone, iPad, and Mac | Mac with Apple Silicon (M1 / M2 / M3 / M4+) |
| Audio leaves your device? | Streamed directly to OpenAI API | Never—processed locally on Mac unified memory |
| AI credit usage | Metered from monthly / pack balance | 0 AI credits used |
| Output modality | Spoken voice + side-by-side text + vision | Live transcript + written text answers |
Two Proven Open-Weights Models Working Together
Rather than relying on a toy model, local Mac mode pairs two dedicated engines tuned for Apple's unified memory architecture: Whisper Large v3 Turbo handles multilingual speech-to-text transcription, and Qwen 3 8B generates contextual written answers alongside your transcript.
Once both model packages are downloaded inside the Mac app (while your subscription is active), inference runs straight on your Mac's GPU and Neural Engine.
When to Pick Local Mode vs Cloud Realtime Mode
Use Local Mac Mode whenever data-residency boundaries prohibit sending spoken audio to external cloud APIs, when working offline on trains or flights, or during multi-hour solo dictation and lecture-review marathons where you want zero credit burn.
Switch back to Cloud Realtime Mode whenever you are on iPhone or iPad, when you need spoken two-way voice interpretation out loud, or when you want frontier multimodal reasoning over attached screenshots.
Hardware Headroom Tips for Smooth Local Inference
An 8-billion-parameter model plus Whisper Large v3 Turbo benefits from clean unified memory headroom. Before starting a long local session on an 8 GB or 16 GB MacBook Air, close idle browser windows with dozens of heavy tabs or Docker containers so macOS keeps model weights resident in fast memory without swap.
Pre-Session Checklist
Saved locally in your browser so you can check your setup before starting.
Frequently Asked Questions
Can iPhone or iPad run Qwen offline too?
No—local Whisper + Qwen execution is exclusive to Apple Silicon Macs. On iPhone and iPad, Apple's built-in Speech recognition can run on-device for transcription and readback, while AI answers require an internet connection.
Do local Mac sessions sync to my iPhone afterward?
No. GPT Realtime Assistant has no cloud sync database for transcripts or notes—every session, transcript, note, and task stays strictly on the device where you created it.
Continue Exploring
Following Fast Meetings and Lectures with Live Transcripts and Instant Answers
Read spoken conversations as they unfold alongside instant AI explanations on iPhone, iPad, and Mac—without missing the thread while taking manual notes.
Read session guide →
Preparing a Live Translation Session for Doctor Visits and Official Appointments
How to prepare vocabulary, two-way voice interpretation, and safety checks on iPhone or iPad when navigating healthcare, school, or municipal appointments abroad.
Read session guide →