Claude's voice mode now runs on the good models, and can send your email
Anthropic opened voice conversations to its Opus and Sonnet models across phone, desktop, and web. It still sounds less natural than ChatGPT or Gemini, but it is the only one that can write and send an email while you talk.
Talking to an AI assistant has mostly been a party trick with a catch: the voice option usually ran on the smallest, fastest model, so the conversation felt quick but the thinking felt thin. Anthropic has now removed that catch. Claude’s voice mode runs on the more capable Opus and Sonnet models, not just the lightweight Haiku, and you can switch between them mid-conversation on the mobile app, the desktop app, and the web.
The mode supports eleven languages and, more usefully, reaches into whatever tools you have connected. If Gmail, Google Calendar, or Slack is linked to your account, you can dictate an email and have Claude compose and send it without touching a keyboard. Anthropic is currently the only major provider where that works straight from a voice conversation.
It is worth being honest about where it still loses. OpenAI’s voice mode uses full-duplex audio, meaning it listens and speaks at the same time, so you can interrupt it the way you would interrupt a person. Google’s Gemini Live works similarly but is limited to phones. Claude uses a turn-based system: it waits for you to finish before it answers. In practice, both OpenAI and Google feel more natural, with smoother speech and better handling of the small collisions of real conversation. Claude’s advantage is not the talking, it is what happens after the talking.
Here is the shift underneath this. For two years, voice assistants competed on sounding human. That race is largely settled and the differences are getting small. The new competition is about what the assistant can actually do once it understands you, which means access to your calendar, your inbox, your files. That is a much less glamorous feature than a natural-sounding voice, and a much more consequential one, because it turns a conversation into an action. It also means the model behind the voice suddenly matters again: a smart model that can send email is a different product from a fast model that can only chat.
A caveat that belongs in the same breath: everything that makes this useful also widens the surface for mistakes. A voice command misheard in a noisy room, or a request that Claude interprets slightly too liberally, now has your outbox attached to it. Keep the confirmation steps on until you trust it.
What this means for you: If you already pay for Claude, this is free and worth ten minutes of trying. The obvious use is hands-free: dictating a reply while walking, asking what is on your calendar, thinking out loud through a problem and getting a real answer rather than a canned one. If you are choosing between assistants purely on conversation quality, ChatGPT and Gemini still feel more fluid. If you want the assistant to do something at the end of the sentence, Claude currently goes further. For most people the honest verdict is: try it once, and keep it if dictation fits how you work.
Sources
Source: https://claude.com/blog/think-through-hard-problems-in-voice-mode
Not everyone believes OpenAI's rogue agent story, and the doubts are worth hearing
OpenAI says its models escaped a test sandbox and hacked Hugging Face. A Cornell researcher argues the disclosure follows a pattern the company has used since 2019: warn that the technology is dangerous, and let investors hear powerful.
