Skip to content

Anthropic

Claude’s web app hides a Voice tab marked Preview

2 min read

Written by an AI model from the linked sources and checked against them by a second model. How we work

Image: testingcatalog.com

Anthropic appears to be testing a separate “Voice” section in Claude’s web app. Testing Catalog reported on October 10 that it had found a hidden navigation item for it, labeled “Preview”.

The label refers to Anthropic’s internal preview environment, so the section isn’t public. The outlet found no details on what the section would do or how it would differ from Claude’s current voice mode, and there’s no sign of when it might ship.

The tab came a few days after a related change in the mobile apps. Anthropic started asking users for permission to train on their voice chats, with a prompt that reads “Allow us to use your voice data to improve our AI models.” The fine print says “audio recordings and voice chat data help improve how Anthropic AI models understand and respond to speech.” This consent is separate from the one for text conversations, and it’s off by default. Testing Catalog reads the data collection as a possible sign that Anthropic is preparing to build its own speech models. It adds that the data could also go toward the voice features Claude already has.

The provider behind Claude’s current voice mode hasn’t been officially confirmed, though Anthropic has held talks with ElevenLabs about a possible voice partnership. Anthropic’s API offers no text-to-speech model. The company also hasn’t released a native end-to-end real-time audio model: a single model that takes in audio and answers in audio, with no separate transcription and speech-synthesis steps around a text model.

In a Threads post, Testing Catalog argued that its own voice models matter to Anthropic if it wants to keep shipping tools for “vibe coders” or to build a proactive assistant. Its write-up suggests a dedicated Voice interface could let people talk to Claude more naturally while it works on a task. The outlet also says it can’t yet tell whether Anthropic is building a real-time voice model, redesigning the interface or testing something else entirely.

The last voice mode update came on July 23. Voice mode had run only on Haiku, and the update let users choose Opus, Sonnet or Haiku. It also gave voice mode access to connected apps such as Gmail, Google Calendar, Slack, Canva and Notion. TechCrunch reported that Anthropic didn’t change the voice model in that release and hasn’t detailed what voice stack it uses. Anthropic said the July release was “focused on intelligence and tool access” and promised “more to share later this year,” TNW reported.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.