The voice assistant can now switch models mid conversation, reach into Gmail, Slack, and Canva, and converse in nine new languages out of beta.
Anthropic has lifted the depth ceiling on Claude's voice mode. The voice assistant used to run only on Haiku, the company's smallest and fastest model. With Opus and Sonnet now behind the same voice interface, a spoken conversation can take on the multi-step, agentic work the lighter model was never tuned for, and the assistant can now act across Gmail, Slack, and Canva.
The change, reported by The Verge from an Anthropic blog post, lets a user swap between Haiku, Sonnet, and Opus mid-conversation to match the task. Anthropic frames the move as a response to subscribers who had been pushing voice into "real business problems" that Haiku could not handle. Concrete examples the company cites include a calendar reshuffled when a train is late and a conversation turned into a one-page pitch.
Voice mode exited beta in French, German, Spanish, Hindi, Indonesian, Italian, Japanese, Korean, and Portuguese, on top of English.
What is still unmeasured is how often the model swap will surface in everyday use. The depth unlock is real; the next test is whether mid-conversation Opus becomes a normal pattern or stays a power-user trick.