Notes take too long on my Mac
Transcription is fast on any Apple-silicon Mac. Notes are the slow part, because by default they come from a language model running entirely on your machine through the built-in engine, and that model needs memory. On computers with 8 GB the engine picks Qwen2.5 1.5B; from 16 GB up it picks Qwen3 4B. Either can be chosen under Settings → AI notes → Model.
What to expect
Measured on 22 September 2026 on an 8 GB M-series Mac:
- With Qwen2.5 1.5B, the model the engine picks on 8 GB: about 35 seconds for a 5-minute clip, so roughly five minutes for a 40-minute meeting.
- With the larger Qwen3 4B model: a 39-minute meeting (377 transcript segments) took about 12 minutes, and an 18-minute meeting about 4 minutes.
- The machine stays responsive throughout. Earlier out-of-memory crashes are fixed.
On a 16 GB Mac the engine picks Qwen3 4B, gives it a larger context window, and runs it roughly three times faster than on 8 GB. The app shows the estimate before it starts, so you’re never guessing.
Three ways to speed it up
1. Use your own cloud key (about 20 seconds). Under Settings → AI notes → Provider, choose Anthropic or an OpenAI-compatible provider and paste your key. Only the transcript text and your template are sent, directly to that provider, and the Privacy Ledger records every request. Audio never leaves. This is the recommended setup on 8 GB machines.
2. Stay with the smaller local model. On 8 GB machines the engine already picks the 1.5B model, which is noticeably faster and still produces usable summaries and action items. If you switched to Qwen3 4B for quality, you can switch back under Settings → AI notes → Model.
3. Turn off auto-notes. With Auto notes off, nothing runs until you click Generate notes on a meeting, so short calls that do not need notes never cost you a wait. This also helps you stay within the 3 notes a month on the Free plan after the trial.
Why two defaults?
Because notes quality matters more than speed for most people, and Qwen3 4B is the smallest model that reliably gets decisions and owners right, so it is the default from 16 GB up. On 8 GB it would take about 12 minutes for a 40-minute meeting, which is why the engine starts with the 1.5B model there and shows you the cost before you switch. You can change it in one click.