Claude Code voice mode rolls out with spacebar push-to-talk

Anthropic is gradually rolling out push-to-talk voice dictation in Claude Code, inserting transcripts right at your cursor. It’s live for about 5% of users now, available across paid plans, and transcription tokens won’t count against rate limits.

Claude Code voice mode rolls out with spacebar push-to-talk

TL;DR

  • Gradual rollout: Claude Code voice mode live for ~5%; ramping over coming weeks; welcome-screen note indicates access
  • Push-to-talk dictation: Hold Space to speak, release to finish; transcript streams at cursor position
  • Inline editing model: Voice input inserts directly into the editor, enabling mixed typing and dictation without switching modes
  • Toggle command: Enable or disable via /voice
  • Metering: No extra cost; transcription tokens excluded from rate limits
  • Availability: Rolling out to Pro, Max, Team, Enterprise; no plans for Agent SDK support currently

Claude Code voice mode is starting a gradual rollout, adding push-to-talk dictation directly inside the coding-focused Claude experience. Thariq (@trq212) says it’s live for about 5% of users today and will ramp “through the coming weeks,” with access indicated via a note on the welcome screen.

Push-to-talk that types where the cursor is

The most developer-friendly detail here is the interaction model: voice input is designed to behave like a quick inline editing tool, not a separate chatty mode.

According to Thariq, using it is straightforward:

  • Hold space, speak, then release (push-to-talk).
  • The transcript streams in at the cursor position.

That cursor-based insertion changes the feel compared to voice features that live in a separate sidebar or modal. The intent seems to be letting prompts take on a more “mixed media” shape: start typing, use voice for the “messy middle,” then keep typing—without clobbering what’s already on the page.

Voice mode can be toggled with /voice.

Pricing, limits, and who gets it

Anthropic’s approach to metering is also notable. Thariq says voice mode doesn’t cost extra, and tokens for voice transcription don’t count against rate limits.

Availability is described as rolling out to:

  • Pro
  • Max
  • Team
  • Enterprise

Early questions: models, SDK support, and “talking back”

A number of replies immediately pressed for implementation details—particularly which model handles transcription and whether it’s the same as the voice mode in the Claude mobile app. Those questions appear in the thread, but aren’t answered in these posts.

There’s also at least one clear boundary: asked whether it will be supported in the Agent SDK, Thariq responded that there are no plans right now (while still calling it “interesting”).

Separately, some commenters asked for voice playback or for Claude Code to “talk back,” but the rollout being described here is specifically voice input and transcription into the editor.

For more, see the original thread: https://x.com/trq212/status/2028628570692890800

Continue the conversation on Slack

Did this article spark your interest? Join our community of experts and enthusiasts to dive deeper, ask questions, and share your ideas.

Join our community