Anthropic has expanded Claude's voice mode to work with its Opus and Sonnet models, the more capable ones in its lineup. Until now, you could only use voice with Haiku, the lightest and fastest of the three. The update, announced on July 23, 2026, also brings voice interactions into Gmail, Slack, and Canva, and adds nine new languages. The Verge

Think of Anthropic's three models as tiers of service. Haiku is the quick, lightweight one — fast to respond but less equipped for complex tasks. Opus and Sonnet are the heavier thinkers, better at nuanced reasoning. Before this update, if you wanted to talk to Claude out loud, you were stuck with Haiku. Now you can have a spoken conversation with the smarter models too.

Two switching features come with the expansion. You can toggle between typing and talking within a single conversation, and you can switch between models mid-conversation without starting over. So you could ask a quick question using Haiku by voice, then switch to Opus when you need deeper thinking, all in the same chat. Anthropic has not explained exactly how it keeps the conversation intact when you jump between models of different sizes, so some technical questions remain open.

The update also brings Claude's voice into Gmail, Slack, and Canva. Instead of opening a separate Claude app, you could speak a request that works on your email, your Slack messages, or a Canva design. This positions Claude's voice not just as a chat feature but as a hands-free tool that works across the apps you already use.

Nine languages have moved out of beta testing and into full availability for voice mode: French, German, Spanish, Hindi, Indonesian, Italian, Japanese, Korean, and Portuguese. Previously, non-English voice support was still in beta. Moving to full availability suggests Anthropic is now confident in the quality of speech recognition and synthesis across these languages, though the company has not published accuracy numbers.

The update covers three things at once: which models you can use by voice, which apps voice works inside, and which languages are supported. Each addresses a real limitation. Opus and Sonnet in voice mode remove the cap that kept voice users on the less capable model. The Gmail, Slack, and Canva integrations put Claude where you already work. The nine-language expansion opens the feature to far more people than English-speaking markets alone.

Looking at the broader landscape, voice has become a standard feature among the major AI companies rather than something that sets one apart. OpenAI, Google, and now Anthropic all offer voice interaction with their best models. What sets them apart now is the quality of the experience, how many apps it connects to, and how flexible it is. Claude's mid-conversation model switching is notable here: it lets you treat the model as a setting you can change on the fly, rather than something locked in when you start. Whether competitors copy this will be worth watching.

The risk for Anthropic is execution. Opus is a much larger model than Haiku, and larger models take longer to produce a response. In text, a few extra seconds is barely noticeable. In voice, a delay before Claude starts speaking can make a smart model feel slow and less useful. How Anthropic handles that tradeoff inside a real-time voice conversation is a question this update raises but does not yet answer.