Voice Mode grew up,
and started reaching your inbox.
On 23 July 2026, Voice Mode expanded from Haiku-only to also support Opus and Sonnet, added mid-conversation model switching, and started reaching connected tools including Gmail and Slack. The capability jump is real. So is the new way it can go wrong if nobody sets a boundary first.
Voice Mode lets you talk to Claude instead of typing. Until 23 July 2026 it ran on Haiku alone; the update that day brought Opus and Sonnet into voice sessions, added the ability to switch models mid-conversation, and connected voice sessions to tools including Gmail and Slack for the first time.
Out of the Haiku-only cage
Before 23 July, Voice Mode ran on Haiku only, fast and cheap for a quick spoken question, but a poor fit for anything needing real reasoning. Most people who tried it for actual work went quietly back to typing. Opening voice access to Opus and Sonnet changes that calculation for the first time. The exception is the free plan, which keeps voice Haiku-only with a single connected app.
Mid-conversation model switching
You can now move between models inside a single voice session without restarting it, start on a fast model for quick back-and-forth, then step up the moment a request gets complicated. Small mechanic, real difference to how a voice session actually feels to use.
Gmail and Slack access changes the stakes
Voice sessions can now reach connected tools including Gmail, Google Calendar and Slack, useful for hands-free triage or dictating a reply between meetings. It also means a spoken instruction can trigger a real action in a way a typed draft usually does not until you hit send yourself.
A genuinely longer language list
English plus, in beta, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese and Spanish, with Spain and Latin America covered separately. For non-English-first teams, this is the point Voice Mode stops being an English-only add-on.
The one rule to set before rollout
Decide, before anyone uses it for real work, which actions Voice Mode can take unsupervised and which need a confirmation step. The Gmail and Slack access makes this a genuine decision rather than a formality, and it is much cheaper to set the boundary now than to explain an unintended send afterwards.
Whether Voice Mode is worth adopting properly, rather than left as a curiosity, depends entirely on how much of your working day already happens in Gmail and Slack. If it is a lot, this update earns a real look, ideally with the confirm-or-autonomous boundary decided in advance rather than discovered live.
Things people
usually ask.
Opus and Sonnet, as of 23 July 2026, alongside the Haiku support it launched with. You can also switch models mid-conversation without restarting the session. The Fable models are not yet supported in voice.
Voice sessions can now reach connected tools including Gmail and Slack. Whether a spoken instruction sends something immediately or waits for confirmation is a setting you choose, not a default you can assume, and it is worth deciding deliberately before rollout rather than discovering it by accident.
English plus, in beta, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese and Spanish in both Spain and Latin American variants, a meaningfully longer list than before the 23 July update.