Claude's voice mode just got a lot smarter

Talking to an AI out loud is great for brainstorming, but until now, most voice bots weren't smart enough to handle real business logic.

Anthropic rolled out a major beta update to Claude's voice mode. Users can now run voice conversations on its heavyweight models, Claude Opus and Claude Sonnet, instead of being stuck with the lightweight Haiku.

Previously, voice mode defaulted to Haiku strictly for speed. It answered fast, but struggled whenever users tried to think through complex pitches, mock interviews, or messy product roadmaps.

Smarter models and real tool actions

With this update, voice mode automatically picks up the last model you used in your text chat and runs its fastest variant. You can also switch between Opus, Sonnet, and Haiku mid-conversation right from the model picker.

Instead of just spitting out quick bullet points, the bigger models are designed to act as sounding boards. Claude will listen, ask probing follow-up questions, and help you find gaps in your thinking before you commit to an idea.

Even better, Claude can now take action using your connected apps. Once you wrap up a spoken brainstorm, you can tell Claude to draft an email in Gmail, push a Google Calendar meeting back by 30 minutes, or turn the pitch notes into a document in Notion or a one-pager in Canva. Claude asks for permission before running any tool action.

Anthropic also added support for 11 languages, including Japanese, Korean, German, Hindi, and Spanish. It does not auto-detect your language yet, so you need to pick it in voice settings or tell Claude out loud to switch.

The feature is available across mobile, desktop, and web. Free users remain limited to Haiku and one connected tool, while paid subscribers get access to the smarter models and all connected apps.

Why it matters

Anthropic and OpenAI are taking wildly different bets on what voice AI should actually do.

OpenAI has spent its energy perfecting native, low-latency conversational presence with expressive accents and slick interruption handling. Anthropic, by contrast, did not even touch its underlying voice stack for this release. Claude still uses traditional turn-taking, meaning it pauses to think before answering, and you cannot smoothly talk over it.

Yet, we think Anthropic's approach is far more practical for serious work right now.

A hyper-realistic voice does not help much if the model lacks deep reasoning or cannot touch your workspace. While OpenAI's voice mode remains essentially isolated from third-party tools, Claude can now turn an unstructured verbal ramble directly into a scheduled meeting, a Slack follow-up, or a drafted deck.

The main friction point will be latency. Opus and Sonnet take longer to deliberate, and without refined interruption controls, verbose voice replies might occasionally test your patience.

Still, if you do your best thinking away from the keyboard, Claude's upgraded voice mode is well worth putting to work today.