Claude voice mode can now turn conversations into real work

Anthropic has expanded Claude’s voice mode to include its more capable Opus and Sonnet models, alongside the faster Haiku model. The update is designed for longer, more complex spoken conversations, such as preparing a client pitch, testing an idea or reviewing a decision.

Users can select a model during a voice conversation or let Claude use the most recent model chosen in text chat. According to Anthropic, voice mode runs the fastest available version of the selected model to keep responses quick. This also allows users to move between speaking and typing without starting a new conversation.

The change shifts voice mode beyond short spoken queries. Anthropic says Claude can ask follow-up questions, challenge gaps in reasoning and help people develop unfinished ideas. Suggested uses include rehearsing important meetings, assessing why a video performed well and brainstorming a product roadmap with competitive research.

Connected tools add actions

Claude voice mode can also access connected services including Gmail, Google Calendar, Slack, Canva and Notion. A user could ask Claude to move a calendar meeting, summarise recent emails and draft replies, or turn a discussion into a one-page document. Anthropic says Claude asks for permission before using a connected tool.

The feature is available in beta on mobile, desktop and web. The company says the phone experience works best. Free users receive Haiku, one connected tool and access to all supported voice languages. Paid subscribers can use additional models and their connected tools. Voice chats count against normal usage limits.

Voice mode now supports English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Brazilian Portuguese and two Spanish variants. Users must select or verbally request a language change, as Claude does not automatically detect the spoken language.

Sources

• Think through hard problems in voice mode – Claude by Anthropic
• Anthropic updates Claude voice mode with more capable models – TechCrunch