Claude voice mode now runs on Anthropic’s Opus and Sonnet models

Anthropic has upgraded Claude voice mode so users can run spoken conversations on its Opus and Sonnet models, rather than only the smaller Haiku model. The change matters because voice interfaces for AI assistants are increasingly being judged less on novelty and more on whether they can handle higher-value work across the same tools people use on desktop and mobile.

The updated mode is available in the Claude mobile app, desktop app and web chat, according to Anthropic. Users can change the underlying model during an active conversation, a useful addition for people who may want a faster model for routine requests and a more capable one for harder prompts without restarting the session.

Anthropic says the voice feature supports 11 languages. It can also connect with external tools, including Gmail, Google Calendar and Slack, if the user grants access. In practice, that means a user can dictate an email, have Claude draft it and send or save it from voice mode, according to the company.

What changed in Claude voice mode?

The main change is model choice. Claude’s voice experience previously ran on Haiku, while the new version adds Anthropic’s more capable Opus and Sonnet models and lets users move between them mid-conversation.

That is a product positioning move as much as a technical one. Voice mode is no longer only a lightweight conversational layer over a chatbot. Anthropic is tying it to its higher-end models and to productivity tools, which pushes the feature toward workflow execution rather than spoken search or casual Q&A.

Anthropic did not disclose new pricing, usage limits or enterprise-specific controls in the details reported by The Decoder. Those omissions matter for teams that might use voice mode in work accounts connected to email, calendars or Slack, where permissioning and auditability can become as important as model quality.

How does Claude voice mode compare with ChatGPT and Gemini?

The Decoder reported that OpenAI’s GPT-Live uses full-duplex audio, meaning it can listen and speak at the same time. Claude’s implementation is turn-based: it waits until the user finishes before answering. Google’s Gemini Live is described as using a similar turn-based approach, though The Decoder said it is limited to smartphones.

That difference affects the feel of the product. The Decoder said OpenAI and Google’s voice products come across as more natural because of stronger speech processing. It also said all three systems handle voice-based search well.

Claude’s differentiator, according to The Decoder, is tool access from audio mode. Anthropic is currently described as the only provider among the three that lets users compose and save an email directly from the voice interface.

For Anthropic, the update brings Claude’s voice feature closer to the rest of its product strategy: fewer standalone demos, more integration into everyday knowledge work. The competitive gap is still visible in real-time speech handling, but the Opus and Sonnet expansion gives Claude users a more capable model path when voice sessions move from simple questions to work that needs stronger reasoning or connected tools.

This story draws on original reporting from The Decoder.