# Claude’s new voice mode isn’t chasing natural. It’s chasing your to-do list. > Source: > Published: 2026-07-24 16:13:08+00:00 Anthropic wants you to talk to Claude, and then watch it act. On Thursday it upgraded Claude’s voice mode to run on its most capable models, Opus and Sonnet, the company [announced](https://claude.com/blog/think-through-hard-problems-in-voice-mode). Voice mode launched last year, and until now it ran only on Haiku, Anthropic’s fastest but shallowest model. That was fine for quick questions. But people started using it to work through real problems out loud, and Haiku kept things quick, Anthropic said, “but not always deep”. The pitch is a smarter sounding board. Opus and Sonnet are built for hard problems. You can talk through a half-formed idea, and Claude will ask follow-up questions and push your thinking rather than just answer. You switch models mid-conversation, and it uses the fastest version of whichever you pick. It defaults to the last model you used in text. Anthropic pitches it for practising a pitch, weighing two job offers, or thinking out loud before a hard call. ## From talking to doing The bigger shift is that voice mode can now [act](https://thenextweb.com/news/1password-claude-credential-zero-exposure-agentic-mode). Connect your apps and you can tell Claude to push a meeting on Google Calendar. It can summarise the day’s emails and draft replies, or turn a spoken pitch into a one-page document in Canva. It can also reach [Slack](https://thenextweb.com/news/block-buzz-humans-ai-agents-workspace) and Notion. Claude asks permission before touching a tool. Voice mode now speaks many more languages too, including French, German, Italian, Japanese and both flavours of Spanish. You have to tell it to switch, though, as it will not detect the change on its own. ## Not chasing natural This is where Anthropic parts from OpenAI. Weeks ago, OpenAI gave ChatGPT a more natural, [full-duplex voice](https://techcrunch.com/2026/07/08/openai-releases-new-voice-models-for-more-natural-live-conversations/) that can listen and talk at once. Claude’s voice mode is turn-based: it listens, pauses to think, then replies. In practice that feels less fluid than ChatGPT. Anthropic did not touch the underlying voice stack, so there are no new tricks like better interruption handling. “This release is focused on intelligence and tool access,” it told [Engadget](https://www.engadget.com/2221938/claude-voice-mode-just-got-smarter/), promising “more to share later this year”. The split is telling. [OpenAI](https://thenextweb.com/news/openai-agents-10-million-chatgpt-work) is making its [assistant](https://thenextweb.com/news/amazon-moonraker-alexa-ai-agent-cost) sound more human. [Anthropic](https://thenextweb.com/news/anthropic-meta-10-billion-compute-deal-talks) is making its assistant do your job. The new voice mode is in beta for everyone on mobile, desktop and web, though it works best on a phone. Free users get Haiku and a single connected app. Everyone else gets the smarter models and the full set of tools, and the promise that the talking is only the start. ## Get the TNW newsletter Get the most important tech news in your inbox each week.