Anthropic has extended Claude’s voice mode to Opus and Sonnet models, no longer limited to Haiku. Now you can speak with the most capable model, switch languages mid-conversation, and let Claude compose emails through Gmail, read Calendar, or write to Slack, all by voice.

The move comes two days after OpenAI’s Presence, the enterprise platform that merges voice and chat into a single agentic product, and the $5 billion AMD-Anthropic agreement for 2 gigawatts of MI450 GPU compute. Three moves in one week converge on a single point: voice becomes the channel through which agents operate.

Why it matters to you. If you’re bringing agents into your work, the infrastructure is solidifying. Model, voice channel, and hardware end up in the same hands. Claude today is the only one that lets you compose and send an email directly from voice. OpenAI and Google have more natural audio (full-duplex versus Claude’s turn-based), but Anthropic bets on integration with your tools.

The limitation is honest: Claude’s voice waits for you to finish speaking before responding, and it sounds less fluid than GPT-Live. For those evaluating which platform to adopt, the choice narrows to a concrete trade-off: naturalness of conversation versus direct control over tools.

In detail

What came before. Claude’s voice mode ran only on Haiku, the lightest model in the family. Fast response times, but limited on tasks requiring extended reasoning or refined writing. Now Opus and Sonnet are available on mobile, desktop, and web apps, and you can switch models mid-conversation if you need more power or speed.

How voice works. Claude uses a turn-based system: you speak, it waits for you to finish, then responds. It’s the same pattern as Google’s Gemini Live. OpenAI’s GPT-Live instead is full-duplex: it speaks and listens simultaneously, like a phone call. In practice, GPT-Live and Gemini sound more natural because there’s no silence between exchanges. Claude compensates with something the others don’t have: direct access to your work tools.

The advantage that counts. Anthropic is currently the only provider that lets you compose and send an email directly from voice mode. If you’ve connected Gmail, Calendar, or Slack, you can dictate a message to Claude and it writes, saves, or sends it without you reaching for the keyboard. It’s the difference between a voice assistant that answers questions and an agent that takes actions on your behalf.

The converging picture. Three moves in one week point in the same direction. OpenAI formalized an enterprise agentic platform for voice and chat. Anthropic signed on for 2 gigawatts of MI450 GPU compute to sustain training and inference. Voice on Opus is the third piece: the most capable model, the voice channel, and the compute to back it, all in the same hands. For those evaluating which platform to adopt, the ecosystem is consolidating into vertical stacks where hardware, model, and interface align.

Where sources diverge. The Decoder notes that Claude loses on voice naturalness compared to OpenAI and Google, but wins on tool integration. It’s a real trade-off, not a clear win. The open question is whether Anthropic will add full-duplex, or whether the bet is that Claude users value direct control over tools more than conversational fluidity.

Limitations. Eleven languages supported, but sources don’t specify which ones or how well less common languages perform. Integration with Gmail, Calendar, and Slack requires explicit permissions: the security pattern is sound, but in team settings you’ll need to define who can connect what and which data the agent can access.