Anthropic has enhanced Claude's voice mode, allowing it to leverage more powerful AI models for improved performance. Previously, voice queries were routed through Haiku, the smallest model, to minimize latency, which sometimes led to struggles with complex requests.

Now, voice mode can access Anthropic's Sonnet and Opus models, providing greater intelligence for conversations. Paid users can switch between Haiku, Sonnet, or Opus mid-conversation and the tool will default to the last system used for text chat. Additionally, voice mode can now integrate with apps like Gmail and Slack, provided users grant permission, to pull context from connected services.

While not fully duplex like OpenAI's GPT-Live, Claude's voice mode uses a turn-based architecture. It also supports additional languages, including Indonesian, though it requires manual language selection for mid-conversation switches. The update is rolling out in beta to all users on desktop and mobile apps, and the web client. Free users will be limited to a single connection and prompts will still use Haiku.