Anthropic overhauled Claude's voice mode on Thursday, letting users route conversations through Opus, Sonnet, or Haiku instead of the Haiku-only stack that shipped last year. The upgrade also plugs voice directly into Gmail, Google Calendar, Slack, Canva, and Notion, closing a gap that has kept spoken AI assistants stuck at the level of read-aloud chatbots. Anthropic is pitching the release as voice for real work — longer conversations, pitch rehearsals, communication-style feedback, product-market brainstorms — rather than quick Q&A.
The model-picker behavior is the mechanical shift. Voice mode now defaults to the fastest variant of whichever Claude model the user last selected in text chat, so a Sonnet-heavy workflow carries into audio without a manual switch. That single change lifts the ceiling on what voice can handle: Haiku is tuned for latency, Sonnet for balanced reasoning, and Opus for complex multi-step tasks. Previously, every spoken query got routed to Haiku regardless of complexity.
The tool integrations are the bigger competitive move. A user can now tell Claude by voice to update a meeting slot in Google Calendar, draft an email in Gmail, or create a document in Notion, and the model executes through connected-app permissions. This is the split with OpenAI's recent voice update, which refreshed conversational style in ChatGPT but still cannot reach into external tools to complete work. For anyone using voice as an actual interface — while driving, while cooking, while walking between meetings — tool access is the difference between a demo and a workflow.
Key facts
- 01Claude voice mode can now run on Opus, Sonnet, or Haiku — previously it was Haiku-only.
- 02Voice mode connects to Gmail, Google Calendar, Slack, Canva, and Notion to draft emails, update meetings, and create docs.
- 03Multilingual support covers 10 languages including English, French, German, Hindi, Japanese, Korean, and Spanish.
- 04Free users are capped at the Haiku model with only one connected app; paid tiers get the full stack.
- 05OpenAI's competing ChatGPT voice mode updated conversational style but cannot use external tools.
Multilingual support, which Anthropic added in beta earlier this year, covers 10 languages: English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Brazilian Portuguese, and Latin American/Spain Spanish. Users have to manually specify the language rather than relying on auto-detection, a friction point Anthropic has not yet addressed. The new voice mode ships in beta across all platforms.
Pricing tiers create a clear split. Free users are restricted to Haiku with a single connected app, which limits voice to lightweight tasks and one integration point — a Gmail-only or Notion-only setup, for example. Paying subscribers get access to Opus and Sonnet routing along with the full app roster. Anthropic has not disclosed how heavily tool-connected voice calls will weigh against usage limits on paid plans.
Notably absent from the release: any change to the underlying voice model itself. Anthropic did not update its speech synthesis or recognition stack and has not detailed what components sit under the voice layer. That means users are unlikely to see the conversational polish OpenAI shipped with its latest ChatGPT voice update — smoother interruption handling, more natural pacing, tighter turn-taking. Anthropic is competing on capability and integration, not on the feel of the conversation.
The strategic logic is straightforward. Claude has built its enterprise position on tool use and agentic workflows, and voice was the last major surface where that positioning didn't apply. Routing voice through the full model family and wiring it to Gmail, Calendar, Slack, Canva, and Notion extends the same story — Claude as the model that actually completes tasks — into a modality where OpenAI has historically had a lead on consumer polish.
The trade-off is real. A voice assistant that can draft an email and update a calendar entry but stumbles on interruption handling will feel less pleasant in casual use than one that talks smoothly but can't do anything. Anthropic is betting that professional users will forgive rougher edges for the ability to actually get things done by speaking, while OpenAI is betting the opposite — that conversational quality drives adoption and tool use can be layered on later.
For the broader AI market, this is where the voice-interface race gets interesting. The frontier labs are no longer arguing about who has the most human-sounding synthesized voice; they're arguing about what a voice interface should do. Anthropic's answer — treat it as another surface for agentic work — sets up a direct comparison with whatever tool-enabled voice product OpenAI ships next. The winner will be whichever lab makes voice-driven task completion reliable enough that users stop reaching for the keyboard.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
