Claude voice mode gets Opus, Sonnet, and your Gmail
Claude voice mode now runs Opus and Sonnet and can reach connected apps like Gmail and Slack. What that changes — and what it still can't do for your business line.
Anthropic updated Claude voice mode today so it can run Opus and Sonnet instead of only Haiku, and reach the apps you've connected — Gmail, Google Calendar, Slack, Notion, Canva. Talking to a model is no longer just dictation with a summary at the end. It's a spoken interface that can write to your systems. That's a different thing, and it deserves different scrutiny.
What actually happened
Per Anthropic's changelog and TechCrunch, voice mode now runs on Opus, Sonnet, or Haiku, switchable mid-conversation from the model picker. It defaults to whichever model you last used in text chat, in that model's fastest version. Connected tools work in voice, with Claude asking permission before it touches one. Eleven languages are supported — including French, German, Hindi, Japanese, Korean, Brazilian Portuguese, and two Spanish variants — but you pick the language manually; there's no auto-detection.
The tiering is the part to note. Free accounts get Haiku, one connected tool, and all languages. Paid plans get the larger models and all your connectors. It's rolling out in beta across desktop, mobile, and web.
One thing did not change: the underlying voice model. TechCrunch points out that unlike OpenAI's recent voice release, Anthropic didn't touch the speech layer, so interruption handling and turn-taking behave the same as before. You're getting a smarter brain behind the same mouth.
Why voice AI matters for your business
Be honest about which problem this solves. This is a personal assistant upgrade, not a customer-facing one. Nobody is going to route their main line into a chat app's voice mode. When a client calls your shop and gets put on hold by an agent that talks over them, the model quality is not what they'll remember — the turn-taking is. That layer is unchanged here.
Where it does earn its keep is the drive home. "Draft a reply to the Henderson quote, pull the numbers from the sheet, put a hold on Thursday at 10." That's real work, done hands-free, against systems you already use. If your team lives in email and calendar, this is a genuine hour back per week.
Which is exactly why the permission model deserves a look before you turn it loose. A voice agent with a Gmail connector can send. The prompt that authorizes it is spoken, in a car, with road noise, by someone who is not reading the confirmation. Decide now which connectors get attached to a voice surface and which stay text-only. Our rule with clients: read tools by voice, write tools by keyboard, until you've watched it work for a month.
And treat the model picker as a cost dial, not a quality setting. Opus for reasoning you'd otherwise sit down to do. Haiku for "what's on my calendar." Most voice traffic is the second kind.
Key takeaways
- Claude voice mode now runs Opus, Sonnet, or Haiku, switchable mid-conversation, defaulting to your last text-chat model
- Connected tools — Gmail, Calendar, Slack, Notion, Canva — are reachable by voice, with a permission prompt per tool
- Free plan is capped at Haiku plus one connector; eleven languages, manually selected, no auto-detect
- The speech layer itself is unchanged, so interruption handling is no better — this is not a customer-facing phone agent
- Attach read-only connectors to voice first; keep send-and-write actions on a screen where someone reads the confirmation
Want a voice agent that actually answers your business line? That's a telephony problem, not a chat-app feature — we build them on infrastructure you own, with the escalation paths and call logs your team needs. See what we've shipped or tell us what your phones are doing to you.
Sources: Anthropic changelog, TechCrunch, Engadget.
- #voice-ai
- #anthropic
- #claude
- #connectors
- #ai-agents
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Pilot Protocol's $4.5M: a directory isn't a standard
A seed-stage AI agent network wants to be where agents discover each other. Useful, but know the difference between an open spec and someone else's front door.
Read itNvidia backs Safe Superintelligence, discloses no terms
Nvidia's Safe Superintelligence partnership names no dollar figure and no term length. Reported at $5B. Here's how to read AI deals that omit the numbers.
Read it