Claude’s Voice Mode Adds Opus, Sonnet and Cross-App Control

Anthropic rebuilt Claude’s voice mode on July 23, 2026, swapping its fast but shallow Haiku model for the same Opus and Sonnet models that power its text chat. The update also wires spoken commands directly into Gmail, Google Calendar, Slack, Canva and Notion, so a voice request can now draft an email or move a meeting instead of just answering a question.

The move doubles as a wager. Anthropic is betting that reasoning and tool access beat raw speech speed, even as OpenAI pushes a full-duplex rival called GPT-Live that listens and talks at the same time. The bet pays off for paying subscribers today. It also opens a new front in the security problem that comes with letting an AI agent act inside a person’s inbox.

Claude’s Voice Mode Finally Gets to Think

Until this week, every spoken request to Claude ran through Haiku, Anthropic’s smallest and fastest model, regardless of what a user actually wanted. Engadget reported that the setup handled simple questions fine but struggled once a request got complicated.

The fix had actually been sitting in plain sight for weeks. A model picker appeared inside Claude’s voice interface roughly three weeks before the announcement, letting users tap Opus or Sonnet. It did nothing. Every session still quietly ran on Claude Haiku 4.5 no matter what was selected, according to TestingCatalog, which spotted the dormant feature before Anthropic flipped it on.

Now the picker actually works. Voice mode defaults to whichever model a user last chose in text chat and automatically runs its fastest available version, matching TechCrunch’s description of how Anthropic wants to balance intelligence against response time. Users can still switch between Haiku, Sonnet and Opus mid-conversation.

The same release folded in language support that had been in beta earlier this year. Claude’s voice mode now works, by voice, in:

• English
• French
• German
• Hindi
• Indonesian
• Italian
• Japanese
• Korean
• Portuguese (Brazilian)
• Spanish (Latin America and Spain)

Notably absent from that list: Mandarin Chinese, Arabic and Russian, three of the world’s most widely spoken languages. Anthropic has not said when, or if, voice support will extend to them.

Anthropic Moves Weeks After OpenAI’s Own Voice Reboot

This didn’t happen in a vacuum. TechCrunch framed the timing bluntly, noting Anthropic is moving weeks after OpenAI rolled out its own new conversational models and updated ChatGPT’s voice mode. The two companies have been trading blows over voice and connectors for almost a year.

1. 2025: Claude’s voice mode launches running only on Haiku, built for quick answers rather than complex work.
2. August 2025: OpenAI connects ChatGPT to Gmail, Google Calendar and Contacts, starting with Pro subscribers before reaching Plus users, according to reporting from Search Engine Journal and BleepingComputer.
3. July 9, 2026: OpenAI introduces ChatGPT Work, an agent mode built to run multi-step scheduling and research tasks unattended.
4. July 23, 2026: Anthropic answers with Opus and Sonnet in Claude’s voice mode, plus its own connectors to Gmail, Calendar, Slack, Canva and Notion.

It isn’t only OpenAI circling this territory. Apple used WWDC 2026 to open its rebuilt Siri platform to outside AI models, an admission that its own assistant needed outside help to do what Claude and ChatGPT are now doing natively. Microsoft, meanwhile, has been building its own seven in-house MAI models aimed at cutting Anthropic and OpenAI costs, a sign every major platform wants to own this layer rather than rent it.

Betting on Depth Over Duplex

Anthropic isn’t chasing OpenAI’s approach. Cryptopolitan described the company as taking a different path from GPT-Live, betting on stronger reasoning and tool access over a fully speech-native system. An Anthropic spokesperson told Engadget that voice mode still uses a turn-based design: Claude listens, pauses to think, then answers.

That’s a real trade-off. GPT-Live can process speech while still speaking, closer to how a phone call actually feels. Claude’s version still takes turns. Anthropic has effectively decided that a slower, smarter assistant beats a faster, shallower one, at least for now. The company has told Engadget that deeper voice architecture work is planned later this year, but this release was built around models and tools, not latency.

Readers weighing which assistant to lean on for everyday reasoning tasks can see how the leading models actually compare in a side-by-side test of GPT-4o, Claude 4 and Gemini 2.5, which is a useful baseline for judging whether Opus and Sonnet earn their keep in voice.

Paid Subscribers Get the Whole App Stack

The upgrade isn’t uniform. Free and paid users get meaningfully different versions of the same feature.

Voice conversations count toward the same usage limits as typed ones, and transcripts land in a user’s regular chat history rather than somewhere separate.

What Can Claude’s Voice Do That ChatGPT’s Still Can’t?

Claude’s voice mode can now trigger actions across Gmail, Calendar, Slack, Canva and Notion by voice. ChatGPT’s voice mode cannot yet touch any connected app at all, even though ChatGPT’s text chat has had Gmail and Calendar connectors since August 2025. That gap is the one place Anthropic is clearly ahead.

OpenAI’s own support documentation is direct about the limitation. Its help center states plainly that voice mode currently does not support apps, even as OpenAI has expanded text-based connectors to Gmail, Google Calendar, Google Contacts and Dropbox over the past year for Plus, Pro, Team and Enterprise users. In other words, ChatGPT can read your calendar while you type. Claude can now act on it while you talk.

It’s a narrow edge, and not a permanent one. OpenAI has shown no hesitation about closing feature gaps quickly, and Anthropic’s own voice stack still hasn’t changed under the hood.

Wiring a Voice Assistant Into Your Inbox Carries Its Own Risk

Every one of these connectors is also an opening. Security researchers have spent 2026 documenting how tool-connected AI systems get manipulated, not through stolen passwords, but through carefully worded text the model mistakes for an instruction.

The OWASP Top 10 for LLM Applications lists prompt injection as its number one risk, warning that it may grant unauthorized access to functions that can execute arbitrary commands once tools are wired into an application. Microsoft’s own security team disclosed two vulnerabilities this year, tracked as CVE-2026-25592 and CVE-2026-26030, in a widely used AI agent framework, both since patched.

Independent researcher Simon Willison has described the underlying pattern as a combination of three ingredients: an agent with access to private data, exposure to untrusted content, and a way to send information back out. Give an assistant email, a calendar and a chat tool all at once, and all three conditions are suddenly in one place. Anthropic has not detailed what safeguards sit between a voice command and an actual Gmail send, beyond requiring users to grant app permissions first.

A Trillion-Dollar Appetite for Consumer Wins

Anthropic has enormous incentive to make voice mode stick. The company closed a $65 billion Series H round on May 28, 2026, at a $965 billion post-money valuation, according to its own newsroom announcement, and filed confidentially for an IPO on June 1, 2026.

This funding will help us serve the historic demand we are experiencing, stay at the research frontier, and bring Claude to more of the places where work happens.

Anthropic said in a statement reported by CNBC after the round made it the most valuable startup in Silicon Valley, ahead of OpenAI’s own reported valuation. The company’s Series H funding announcement also disclosed that run-rate revenue had crossed $47 billion earlier that month, up from a $183 billion valuation less than a year before.

A voice mode that can draft your emails and reschedule your meetings is exactly the kind of everyday habit that turns a chatbot subscription into something harder to cancel. Anthropic has said fuller, faster voice architecture is still coming later this year.

Frequently Asked Questions

What happens to a Claude voice conversation afterward?

Transcripts land in the same chat history as a user’s typed conversations, and voice sessions count against the same usage limits as text chats. There is no separate voice-only history or quota.

How do you change the spoken language in Claude?

Language is set manually under Settings, then General, then Voice, then Language, because switching a device’s display language does not automatically change Claude’s spoken language. Users can also ask Claude to switch languages mid-conversation.

Is the new voice mode available to everyone right now?

It’s rolling out in beta across mobile, desktop and web apps for Free, Pro, Max, Team and Enterprise plans, though free accounts stay locked to the Haiku model and a single connected app.

Will Claude’s voice mode ever work in real time, like a phone call?

Not yet. Anthropic has told Engadget that this release prioritized intelligence and tool access over rebuilding the underlying voice stack, with additional voice investment planned later in 2026.