For about a year, talking to Claude meant talking to its smallest brain. Voice mode shipped running exclusively on Haiku, the model built for speed rather than depth. Great for "what's the weather," considerably less great for "help me pressure-test this pitch before Monday." Anyone who tried the second kind of question tended to go back to the keyboard.
That ceiling came down on Thursday, July 23, 2026. Anthropic has opened voice mode to its full model lineup: Opus and Sonnet now sit alongside Haiku, and users can swap between them mid-conversation from a model picker without dropping the thread. Voice mode also inherits whichever model you last used in text chat, and automatically selects the current generation of it, so the choice you already made in typing carries over into speech.
There's a sensible piece of engineering underneath: whichever tier you pick, voice mode runs the fastest available version of it, which is how a reasoning-heavy model avoids turning every reply into an awkward pause.
The second half of the update may matter more than the first. Voice mode can now reach connected apps, among them Gmail, Google Calendar, Google Docs, Slack, Notion and Canva. That turns a spoken conversation into something closer to a command line for your working day. Ask it to scan a calendar for conflicts, summarise an email thread out loud, or draft and save a reply, and it can act rather than explain how you might do it yourself.
That last capability is a genuine differentiator. Among the three big consumer assistants, Claude is currently the only one that will compose and send an email by voice alone. Rivals are still largely narrating instructions back at you.
| Feature | Before | After |
|---|---|---|
| Models available | Haiku only | Haiku, Sonnet, Opus |
| Switch models mid-chat | No | Yes, via model picker |
| Connected app access | None | Gmail, Calendar, Docs, Slack, Notion, Canva |
| Languages | English at launch; multilingual beta added this year | 11 supported, switchable mid-conversation |
| Platforms | Mobile | Mobile, desktop and web (public beta) |
| Free plan | Haiku | Haiku, one connected app |
Eleven languages are now supported: English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Brazilian Portuguese, and Spanish for both Latin America and Spain. Users can move between them without restarting the session.
For bilingual households and cross-border teams, that is the most quietly useful line in the release notes.
Free-tier users get the update too, with limits: Haiku only, and a single connected app. Paid plans unlock the full model range and multiple connections. Voice conversations count against standard usage quotas, and the company's newest Fable model is not yet available in voice mode.
Anthropic was unusually direct about the scope here, describing the release as focused on intelligence and tool access rather than on the voice layer itself. The underlying speech recognition and interruption handling are untouched.
Which means the architectural gap with competitors remains. Claude's voice mode is turn-based: it listens, pauses to think, then speaks. A rival's newest system runs full-duplex, processing speech and generating output simultaneously, and a third major assistant takes a similar turn-based approach but is limited to phones. In side-by-side use, the full-duplex systems still feel more fluid. Anthropic's bet is that most users will trade a beat of latency for an assistant that can actually finish the task. The company has signalled more voice work is coming later this year.
Response times for Opus and Sonnet in voice mode haven't been published, so real-world smoothness is something users will have to judge for themselves.
Anthropic's timing isn't a coincidence. Voice assistants have become one of the fastest-moving areas in AI, with nearly every major company racing to make conversations feel more useful instead of simply more natural.
For years, voice assistants were mostly shortcuts. They could answer a quick question, set a reminder, or tell you tomorrow's weather, but anything more complicated usually meant reaching for a keyboard. That expectation is beginning to change.
The latest generation of AI assistants is being designed to do more than talk. With permission, they can search your files, check your calendar, read documents, summarize emails, and help complete everyday tasks across multiple apps. The competition is no longer about who has the most realistic voice—it's about who can save users the most time.
Claude's latest update fits squarely into that shift.
The biggest change isn't that Claude now speaks through a more capable model. It's that voice is no longer being treated as a lightweight version of the chatbot.
By bringing Opus and Sonnet into voice mode and connecting them with everyday productivity tools, Anthropic is betting that people will increasingly choose to speak to AI when they're working—not just when they're asking simple questions.
Whether that becomes the new normal will depend on the details users notice most: how quickly Claude responds, how reliably it handles connected apps, and whether those conversations genuinely save time instead of creating extra work.
The technology is moving quickly, but one thing hasn't changed. AI assistants are still capable of making mistakes, especially when summarizing information or completing tasks automatically. For anything important, a quick human review remains the safest final step.
Comments