/voiceClaude Code ships push-to-talk dictation. It is free, it is good, and for one specific job it is the right tool. This page is about where that job ends — written by the people who sell the other one, so read it with that in mind and check the links.
/voice actually isRun /voice in the Claude Code CLI and hold Space to record — or tap once to start and
again to send. Your speech appears in the prompt as you speak. It also works in the VS Code extension, and
transcription is tuned for coding vocabulary: regex, OAuth, JSON, and
your project and branch names are fed in as hints. It costs nothing and does not consume tokens.
Three details decide most of the comparison, and all three are in Anthropic's own documentation:
| Claude Code /voice | VocalCode | |
|---|---|---|
| Where the text lands | Claude Code CLI + VS Code extension | Any app that takes a keystroke |
| Where speech is recognised | Anthropic's servers | Your own CPU |
| Account required | Claude.ai account | None |
| Offline | No | Yes |
| Remote / SSH / web | Not supported | Runs where your keyboard is |
| Chinese | Not supported — falls back to English | Native (Paraformer) |
| Price | Free | $4.99 once |
Claude Code's dictation supports twenty languages. Japanese and Korean are among them. Chinese is not. And an unsupported language does not simply refuse — the documentation is explicit about what happens instead:
“If your language setting is not in the supported list, /voice
warns you on enable and falls back to English for dictation.”
Speech decoded by the wrong language does not come out empty; it comes out as confident nonsense — English words that happen to sound like what you said. We know the failure exactly, because we shipped it ourselves once: an early VocalCode guessed the language from the OS locale, and Chinese speakers on English-locale machines had their speech force-decoded as English. We tore that out and made the language a question the user answers, which is why VocalCode now asks once, on first run, and never changes its mind afterwards.
Chinese here runs on Paraformer, a Chinese-first model, not a multilingual one with Chinese bolted on — and it handles English inside a Chinese sentence, which is how most bilingual developers actually talk.
What we do not cover, so you hear it here rather than after paying: Korean, Japanese and Thai are in neither of our models. Claude Code's dictation lists Korean and Japanese and we do not — if those are your languages, that row goes to them, not to us.
One caveat that applies to both tools and is worth understanding before you test either: a dictation engine decodes with the language it was told to use. Point it at the wrong one and you do not get an error, you get the sounds you made spelled out in the wrong alphabet — 안녕하세요 comes back as “Ani Haseyo”, สวัสดี as “Savatika”, Bonjour as “Bangzhou”. That is the same failure whichever product produces it, which is exactly why this one asks you the question on first run instead of guessing.
/voice if……you dictate into Claude Code and nowhere else, you are signed in with a Claude.ai account, and cloud transcription is fine with you. It is free, it is built in, and it is tuned for exactly that prompt box. We are not going to pretend otherwise.
/voice is unavailableThese are not really competitors. Anthropic built a microphone into their own prompt box; we built one for the rest of the operating system. If Claude Code is the only place you ever speak, save your $4.99 — and if it is not, the trial is 30 days and does not ask for a card.