Which Model Does Jarvis Use for Summarizing Meetings?

For meeting and transcript summaries, Jarvis (getjarvis.eu) routes to a long-context model when the transcript is long, since its large context window holds a full hour of dialogue, and to a frontier model when you need clean action items and accurate attribution of who said what. Jarvis is the screen-aware desktop AI assistant for macOS, Windows, and Linux; press Cmd+/ (macOS) or Ctrl+/ (Windows) over a Google Docs transcript, a Notion page, or a recap email and ask for a summary plus next steps. With Notion, Linear, Todoist, and Google Calendar connected, it can also turn decisions into tasks. Routing is automatic at $16/month on Pro, so you get all three models instead of paying for ChatGPT and Claude separately. Jarvis does not record or join calls itself, so you need a transcript from a notetaker first.

Summarizing a meeting is really two jobs, and they favor different models. Holding the whole transcript in memory rewards a long-context engine, which is why Jarvis sends hour-long or multi-speaker logs to a frontier model. Extracting decisions, owners, and deadlines without inventing them rewards careful instruction-following and attribution, which is where a frontier model shines. Jarvis reads the length and the ask, then routes accordingly, so a quick standup recap and a 90-minute strategy call do not get the same treatment. You can always ask it to redo the action items with a different model if the first pass misses a commitment.

The reason to do this in Jarvis rather than paste a transcript into ChatGPT is what happens after the summary. Because Notion, Linear, Todoist, ClickUp, Asana, and Google Calendar are connected, Jarvis can summarize the call and then create the follow-up tickets, add calendar holds, or drop notes into the right Notion page, all from the floating bar. Open the transcript wherever it lives, Google Docs, an Obsidian note, an email recap, press Cmd+/, and say summarize this and file the action items. A standalone model in a browser tab can only hand back text you then re-key into your tools yourself.

AI models