Claude's Voice Mode Now Runs on Opus and Sonnet, Not Just Haiku

Anthropic pushed a significant upgrade to Claude's voice mode on July 23, letting it run on the company's more capable Opus and Sonnet models instead of relying solely on the lightweight Haiku model that powered it at launch last year.

The update also widens what voice mode can actually do. Beyond existing hooks, Claude can now connect to Gmail, Google Calendar, and Slack while talking, and it adds support for Canva and Notion as well.

One capability stands out: Claude can compose and send emails entirely by voice. According to reporting from TechCrunch and The Decoder, neither ChatGPT nor Gemini currently offers that specific function, giving Anthropic a narrow but concrete edge in voice-driven productivity tasks.

Users aren't locked into one model for a whole conversation. Claude now automatically selects a model based on whichever one was most recently used in text chat, and people can manually switch between Opus, Sonnet, and Haiku, or toggle between voice and text, mid-conversation.

Wider language and platform reach

The feature is rolling out in beta across mobile, desktop, and web chat simultaneously, rather than staggering across platforms. Language support has also expanded to eleven options, covering English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese, and Spanish.

Coverage differs slightly on the exact Spanish variant: TechCrunch and The Decoder list "Spanish" as one of the eleven, while AIBusiness separates "Spanish" and "Latin American Spanish" as distinct entries within its own count of eleven. The discrepancy is minor and doesn't change the substance of the expansion.

Free tier stays limited

Access isn't uniform across plans. Free users are capped at the Haiku model and a single connected third-party app, while paid subscribers get the full roster of models and all available integrations.

That tiered structure suggests Anthropic is using voice mode's more advanced capabilities, particularly the Opus and Sonnet access and multi-tool connectivity, as a lever to push free users toward paid plans.

A different architecture than rivals

There's also a technical distinction worth flagging. OpenAI's GPT-Live voice system uses full-duplex audio, allowing simultaneous speaking and listening, whereas Claude's voice mode remains turn-based, meaning users and the assistant speak in sequence rather than overlapping.

That architectural choice may explain why Anthropic's rollout emphasizes tool integration and email drafting rather than framing this as a leap in conversational fluidity. The bet appears to be that task completion, not real-time naturalism, is the more valuable differentiator for professional users.