Anthropic is shipping native voice input inside Claude Code, its command-line AI coding tool. Boris Cherny, the tool's creator, announced the rollout on X, confirming it's live for roughly 5% of users today with a broader ramp over the coming weeks.
The interaction is push-to-talk: hold the spacebar, speak, release. Transcribed text streams directly to the cursor position in real time. That means you can type half a prompt, dictate the messy middle by voice, then finish typing. No overwriting, no replacement. It slots into the existing workflow rather than taking it over.
Cherny called it "our most capable system yet" for voice, though the more interesting detail is the economics. Transcription tokens are free and don't count against rate limits, per reporting from 36Kr. Users who get access will see a note on the Claude Code welcome screen; /voice toggles the feature on and off.
The timing is notable. OpenAI's Codex added a nearly identical hold-spacebar-to-transcribe feature in version 0.105.0, using the Wispr voice engine. Both companies arrived at the same conclusion independently: for describing bugs and architecture decisions, talking is 3-4x faster than typing. The practical use case isn't replacing keyboards but complementing them, especially for natural-language-heavy prompts where precise variable names still need to be typed.
Voice mode is available on Pro, Max, Team, and Enterprise plans. No date yet for general availability beyond the initial 5% cohort.
Bottom Line
Claude Code's native voice mode is free to use, doesn't eat into rate limits, and is rolling out to all paid plan tiers over the coming weeks.
Quick Facts
- Live for ~5% of Claude Code users today
- Push-to-talk: hold spacebar to speak, release to transcribe
- Transcription tokens are free, no rate limit impact
- Toggle with /voice command
- Available on Pro, Max, Team, and Enterprise plans
- OpenAI Codex shipped similar feature in v0.105.0
