Anthropic has quietly rolled out a significant update to Claude’s voice mode, swapping in more powerful models and adding the ability to pull information from apps like Gmail and Slack. The feature, available since last year, lets users speak their prompts instead of typing. But early adopters found it hit-or-miss, especially for complex tasks, because queries were routed through Haiku—Anthropic’s smallest model—to keep latency low. That trade-off meant simple questions worked fine, but anything more nuanced could leave users frustrated.

Starting today, voice mode can tap into the company’s Sonnet and Opus models when needed. For paying Claude users, the tool defaults to whichever model you last used for text chat. You can also switch between Haiku, Sonnet, or Opus mid-conversation using the model picker. Anthropic says voice mode uses the fastest version of the selected model to keep the conversation smooth. Additionally, if you grant permission, Claude can now pull context from connected apps like Gmail and Slack, making the voice experience more useful for real-world tasks.

The update comes with some limitations. Voice mode uses a turn-based architecture: Claude listens, pauses to think, then responds. That means it’s not fully duplex like OpenAI’s GPT-Live, which can process speech and generate output simultaneously—making conversations with Claude feel less natural. The system also can’t automatically detect a language switch mid-conversation; you have to tell it out loud or select the new language in settings. On the plus side, Anthropic has added support for Indonesian and other languages. The new voice mode is rolling out in beta across desktop, mobile, and web. Free accounts are limited to one connection and only Haiku, but can use all supported languages.

Source: Engadget