Lilith Lilith.

Anthropic is extending Claude Voice from the faster Haiku model to Opus and Sonnet, with access to applications including Gmail, Slack and Canva. Model availability follows the user’s plan, and the Free plan allows one connected tool while paid plans allow more. Voice is moving from short questions toward tasks that use business context and can alter a user's next steps.

Sonnet and Opus take on harder voice tasks

The Verge reports that Voice Mode was previously available only on Haiku. Anthropic found that people were also using voice for business problems that the faster, less powerful model was not designed to handle. It is therefore adding stronger models to the same interface.

Users can move between text and voice within one conversation and switch models when needed. Voice is not a separate feature in that setup. It becomes an input to the same context where Claude handles longer and more complex work.

Gmail and Slack turn conversation into a work interface

The Verge gives examples such as converting a conversation into a one page pitch or moving appointments when a user is delayed. Connections to Gmail, Slack and Canva create value beyond dictation alone.

For companies, this offers a practical way to work without a keyboard. It also makes the boundary between suggestion and action less obvious. The system needs to show which application it is reading, what it intends to change and when it is waiting for confirmation.

Natural dialogue can hide the weight of approval

A poor Haiku answer can be corrected with another question. A mistake by a model working with email, Slack or a calendar can affect colleagues and external recipients. Voice also accelerates consent because a spoken sentence feels less formal than a dialog box.

Anthropic is adding French, German, Spanish, Hindi, Indonesian, Italian, Japanese, Korean and Portuguese. Czech is absent from the listed languages, limiting local usefulness.

Enterprise adoption will depend on confirmation and audit trails

Granular permissions, a record of completed steps and required confirmation before sending or changing a calendar will determine adoption. Differences in latency and error rates across Haiku, Sonnet and Opus will matter as well. Voice will earn trust when users can recognize the moment Claude stops speaking and starts acting.

Lilith's verdict

A traveler can dictate a calendar change before opening a laptop. Claude still needs to display every external action clearly before execution, because a natural voice is not consent for unlimited agency.

I keep the external link at the end. First, a concise explanation here — no hunting across someone else's site.

Original source ↗