Claude voice gets Opus and Sonnet as Anthropic challenges GPT-Live

Anthropic is giving Claude’s voice mode a major upgrade, allowing paid users to run spoken conversations on its more powerful Opus and Sonnet models for the first time.

Until now, every voice request was routed through Haiku 4.5, Claude’s smaller and faster model. The upgrade will make voice mode more suitable for handling difficult tasks and is just weeks after OpenAI’s release of their own conversational platform upgrade called GPT-Live.

Claude voice finally moves beyond Haiku

Voice mode has been part of Claude since last year, letting people speak to the chatbot rather than type. The catch, as Engadget noted, was that Anthropic pushed those spoken requests through Haiku, its smallest and fastest model, to keep response times low. That trade worked for quick questions and fell apart on anything harder.

The new release lets users pick Opus or Sonnet instead. Anthropic told TechCrunch that voice now defaults to whichever model the user last used in text chat, running its fastest variant so the exchange still flows. A model picker inside the interface lets users swap between Haiku, Sonnet, and Opus without leaving the conversation, per 9to5Mac.

TestingCatalog, which spotted the feature under a flag before the announcement, reported that the picker had sat inside voice mode for roughly three weeks as a cosmetic control, with every session still routed to Claude Haiku 4.5 regardless of the choice.

Selecting Opus or Sonnet only began changing the actual model this week.

App connectors make voice useful for work tasks

The upgrade also opens voice mode to Claude’s app connectors. TechCrunch reported that the assistant can now reach into Gmail, Google Calendar, Slack, Canva, and Notion, so a spoken request can move a meeting, draft an email, or spin up a Notion document. Engadget confirmed voice can pull context from connected apps once a user grants permission.

Anthropic framed the tasks this unlocks as longer, messier ones: rehearsing a client pitch, getting feedback on how you communicate, or working through product research aloud. “This release is focused on intelligence and tool access,” the company said, adding it plans more voice work later this year.

As Cryptopolitan reported, Anthropic reached a roughly $350 billion valuation in November 2025 and derives approximately 80% of its revenue from more than 300,000 business and enterprise customers.

That makes connector access the more important part of the voice upgrade. Enterprise users can ask Claude questions regarding their emails, calendars, Slack messages, and documents using voice queries, making this tool more relevant to how larger customers have deployed Claude for themselves, as seen in applications like the autonomous agents used in Alberta, spanning 27 ministries, and at Deloitte, with 470,000 employees.

Anthropic expands language support for voice users

9to5Mac listed 11 supported languages: English, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Brazilian Portuguese, and Spanish for both Latin America and Spain.

Moving between these languages does not happen automatically. Engadget said that users would need to say out loud that they are going to switch their language or choose a new one in the voice settings menu. Free accounts remain capped at Haiku and a single connected app but can still speak in every supported language.

Claude and GPT-Live take different paths

These two companies seem to be going about voice technology in different ways. According to a representative from Anthropic, Claude has a turn-based system, meaning it listens, thinks, and then speaks. This is unlike OpenAI’s GPT-Live, which simultaneously listens and speaks. It was released for all ChatGPT subscribers on July 8, 2026, and was referred to as “the primary interface to computing.”

Anthropic did not change Claude’s underlying voice model in this release and has not fully detailed the technology behind its voice stack, TechCrunch reported. That means users may not see major improvements in areas such as smoother interruptions.

According to TestingCatalog, Claude still seems to use text-to-speech technology for its voice, with the voice output coming from ElevenLabs. Nevertheless, TestingCatalog was able to see that interruption processing is still working fine.

One edge Anthropic keeps is tooling. TechCrunch pointed out that OpenAI’s refreshed voice mode changed its conversational style but still cannot use tools to complete tasks.

The new voice mode is rolling out in beta to all users on desktop, mobile, and web.