Claude Code Voice Mode Is Here: Speak Your Code into Existence
March 29, 2026 · 5 min read
TL;DR
Anthropic launched voice mode for Claude Code on March 3, 2026. Activate with /voice — then hold spacebar to speak, release to send. Supports 20 languages, simultaneous typing + speaking, and customizable push-to-talk keybinding. Currently in gradual rollout (5% initially), expanding through April 2026. Free on Pro/Max/Team/Enterprise — no extra charge. Cloud-based transcription: audio sent to Anthropic servers. Not recommended for air-gapped or classified codebases.
How to enable and use voice mode
Update Claude Code
Run: npm update -g @anthropic-ai/claude-code. Voice mode requires v2.1.70+. Windows support was resolved in this version.
Start a Claude Code session
Open your terminal and start a session in your project directory as usual.
Type /voice to activate
Type /voice and press Enter. Voice mode activates immediately.
Hold spacebar to speak
Hold the spacebar while speaking. Your speech is transcribed in real time. Release spacebar to send to Claude.
Type and speak simultaneously
You can paste code, file paths, or error messages in the text field while speaking context verbally — both inputs combine into a single prompt.
Optional: auto-enable on startup
Add voiceEnabled: true to your keybindings.json to activate voice mode automatically in every session. Rebind the push-to-talk key to any key other than spacebar if needed.
5 real use cases voice mode enables
Describe complex logic verbally
Instead of typing a 3-paragraph explanation of a bug, speak it: 'The issue is in the authentication middleware — when the JWT expires during a long-running request, the token refresh fails silently and returns a 200 with empty data instead of a 401.' Faster than typing, often more precise because you describe it as you would to a colleague.
Navigate large codebases hands-free
While scrolling through code with your mouse: 'Find all the places where we call the payment processor API and show me if any of them are missing error handling.' Your hands stay on the mouse while your voice directs the search.
Pair programming dictation flow
Run the code, see an error, speak the error aloud: 'I'm getting a TypeError on line 142 — undefined is not a function — but I know the function exists because I can see it above.' Voice captures the conversational troubleshooting flow more naturally than text.
Architecture planning while typing
Use simultaneous input: speak the high-level intent while pasting file paths in the text input. 'I want to refactor this [paste file path] to use the repository pattern, similar to [paste second file path] but adapted for async operations.' Voice and typing combine without interrupting each other.
Document as you build
Generate documentation verbally as you finish each function: 'Write a JSDoc comment for this function. It takes a user ID and optional date range, queries the database for activity logs, and returns a paginated result set. Throw an error if the user doesn't exist.' Faster than typing full documentation from scratch.
Claude Code voice vs. OpenAI Codex voice input
| Feature | Claude Code Voice | OpenAI Codex Voice |
|---|---|---|
| Launch date | March 3, 2026 | February 25, 2026 |
| Activation | /voice command in terminal | Microphone icon in Codex UI |
| Input method | Push-to-talk (spacebar hold) | Continuous listening toggle |
| Simultaneous typing + speaking | Yes — native support | No — switch between modes |
| Language support | 20 languages | Multiple (count unconfirmed) |
| Transcription | Cloud (Anthropic servers) | Cloud (OpenAI servers) |
| Extra cost | None (included in plan) | None (included in plan) |
| Platform | Windows, macOS, Linux | Web interface, macOS app |
Claude Code's push-to-talk and simultaneous text+voice input are the key differentiators — Codex requires switching between input modes.
Frequently asked questions
How do I enable voice mode in Claude Code?
To enable voice mode in Claude Code (as of March 2026): (1) Update to the latest version: run 'npm update -g @anthropic-ai/claude-code' in your terminal. Voice mode requires v2.1.70 or later. (2) In an active Claude Code session, type '/voice' and press Enter. (3) Voice mode activates. Hold the spacebar to speak — your speech is transcribed in real time. Release the spacebar to send the transcribed input to Claude. (4) To deactivate voice mode, type '/voice' again. To set voice mode to start automatically in every session, add 'voiceEnabled: true' to your Claude Code settings (keybindings.json). The spacebar push-to-talk binding can be rebound to any key to prevent accidental triggers.
Is Claude Code voice mode free?
Claude Code voice mode is included at no additional cost for Pro, Max, Team, and Enterprise subscribers as of March 2026. There is no separate voice tier or per-minute transcription charge. The voice transcription uses Anthropic's cloud-based transcription service, which counts against your standard plan's usage. Claude Code free tier access is limited — voice mode requires at minimum a Pro plan ($20/month for Claude.ai). If you access Claude Code through the API directly, voice transcription is handled client-side and does not affect API token usage directly.
What languages does Claude Code voice mode support?
Claude Code voice mode supports 20 languages as of the March 2026 launch, including: English, Spanish, French, German, Portuguese, Italian, Japanese, Korean, Chinese (Mandarin), Russian, Polish, Turkish, Dutch, Swedish, Norwegian, Danish, Finnish, Czech, Romanian, and Ukrainian. Language detection is automatic — the transcription engine identifies your spoken language and transcribes accordingly without requiring manual language selection. Technical terms, code-specific vocabulary (function names, variable names, file paths), and mixed-language input (e.g., English code terminology spoken while using a non-English interface language) are handled correctly in most cases.
Is it safe to use Claude Code voice mode with proprietary code?
Claude Code voice mode uses cloud-based transcription, meaning your spoken audio is transmitted to Anthropic's servers for processing. This is the primary privacy consideration for users working with proprietary, confidential, or security-sensitive codebases. For most commercial development work on non-classified projects: the privacy risk is equivalent to using Claude Code's text mode, where your code context is already sent to Anthropic's servers. The same data handling terms apply to voice transcription as to text input. For air-gapped environments, classified codebases, or highly sensitive proprietary systems: voice mode is not recommended, as it adds an additional audio transmission layer to the existing text context transmission. Anthropic states that Claude Code conversation data is not used to train models by default for Team and Enterprise plans.
Claude coding skills via Happycapy
Happycapy's Claude Code skill gives you Claude Sonnet 4.6 for coding tasks with persistent project memory, Capymail delivery, and 150+ additional skills — without managing a separate CLI. $17/month. Free tier available.
Try Happycapy Free →