Claude Voice gets Opus and Sonnet, and voice moves into work

Anthropic is extending Claude Voice from Haiku to Opus and Sonnet, with access to apps such as Gmail, Slack and Canva. According to The Verge, Voice Mode had previously been available only on Haiku, Anthropic's faster but less powerful model, while Opus and Sonnet are meant for deeper problem solving.

Voice is no longer just a quick answer channel

Claude Voice started as an interface for quick questions with low latency. Anthropic now says people were using it for business problems that Haiku was not really designed to handle. That is why Sonnet and Opus are coming to voice.

Users can switch between text and voice inside the same conversation and change models as needed. That small workflow detail matters: voice stops being a separate gimmick and becomes another input into the same working context.

Gmail and Slack access change both value and risk

The most interesting part is not speaking to a model. It is connecting voice mode to apps such as Gmail, Slack and Canva. The Verge gives examples such as turning a conversation into a one page pitch or shifting calendar appointments when a train is running late.

For product and enterprise teams, that moves adoption into a different category. Voice UI is convenient, but the value appears when the model understands work context and can act. That also raises the usual hard questions: permissions, audit trails and the boundary between suggestion and action.

A stronger voice model makes mistakes more expensive

If Haiku gives a shallow answer, the user can simply ask again. If Opus or Sonnet works with email, Slack and a calendar, a bad interpretation has more consequences. Voice also makes approval feel casual, because conversation feels less formal than a form.

Anthropic is also expanding voice support to French, German, Spanish, Hindi, Indonesian, Italian, Japanese, Korean and Portuguese. Czech is not on that list, which matters more for local adoption than any broad claim about global reach.

Permission design will decide the everyday value

The next signal is not how many people chat with Claude. It is approval quality: whether users can see what the model is about to do in Gmail or Slack, who can audit it later and where an action stops.

If Anthropic turns voice into a safe entry point for workplace tools, Claude gets closer to a real work agent. If not, this remains a polished dictaphone with too much trust attached.

Lilith's verdict

A voice assistant is harmless while it only answers. Once it reaches into calendars and Slack, a new coworker is in the room, speaking faster than you can read the receipt.