Direct answer — Does VS Code’s built-in voice dictation replace a dedicated dictation app? Not outside VS Code. Version 1.131, released on 29 July 2026, added experimental on-device dictation to chat inputs, text editors and the integrated terminal, using a local Nemotron model. It stops at the edge of the VS Code window. A system-wide app such as Weesper Neon Flow dictates into Slack, email, a browser and any other editor through one hotkey.
For years, VS Code voice dictation meant installing the VS Code Speech extension and hoping it stayed maintained. That changed on 29 July 2026, when VS Code 1.131 shipped built-in dictation that requires no extension, runs an on-device model, and works in chat inputs, text editors and the integrated terminal.
It is a genuinely useful feature, and it comes with the editor you already have. It is also, by design, bounded by the VS Code window.
This guide covers what Microsoft actually shipped, where that boundary sits, and how the VS Code dictation vs Weesper decision plays out for a developer whose day is not spent entirely inside one application. If you are already dictating prompts to a coding agent, this is the comparison to read before you change your setup.
What did VS Code 1.131 actually ship?
Built-in, on-device dictation across three surfaces — chat inputs, text editors and the integrated terminal — released on 29 July 2026 and flagged experimental.
The release notes present it as “Built-in dictation across VS Code (Experimental)”, with “a single speech session and microphone selection shared across all three” surfaces. No extension to install, no account, no API key.
Transcription runs locally. The documentation states that “the default speech recognition model processes microphone audio on your device”, and the release notes add that the model “downloads on first use and keeps audio on your device”.
The default engine is nemotron-3.5-asr-streaming-0.6b, changeable through the dictation.model setting. Three other settings matter: dictation.enabled, dictation.showTranscript and dictation.experimental.llmCleanup.
How you trigger it:
- Chat and the Agents window:
⌘Ion macOS,Ctrl+Ion Windows and Linux - Editor dictation:
⌥⌘Von macOS,Ctrl+Alt+Von Windows and Linux - Integrated terminal: through the Command Palette only
One caveat deserves a second read before you switch it on. dictation.experimental.llmCleanup tidies rough dictation into cleaner text, and the documentation is explicit about the trade: “When dictation.experimental.llmCleanup is enabled, VS Code sends the transcript text, but not the audio, to a Copilot language model for cleanup.” The recording stays on your machine. The words you dictated do not.
Which platforms and languages does it cover?
Windows on x64 and Arm64, macOS on Apple silicon, and Linux on x64 and Arm64 with glibc 2.34 or later. Remote workspaces are covered too.
The documentation lists what is not supported: VS Code for the Web, Intel-based Macs, 32-bit and Arm32 systems, and musl-based Linux distributions such as Alpine. On those, the VS Code Speech extension remains the fallback.
Languages are the vaguer part. VS Code publishes no language list and says only that “Support for more platforms and languages is a work in progress.”
The default model’s own card is more precise about itself: NVIDIA states that nemotron-3.5-asr-streaming-0.6b “transcribes 40 language-locales from one model”, which NVIDIA splits into 19 transcription-ready locales, 13 broad-coverage and 8 adaptation-ready. That figure describes the model as its publisher measured it — it is not a commitment about what the editor integration exposes today.
Where does VS Code’s built-in dictation stop working?
At the edge of the VS Code window. The feature is bound to chat inputs, editors and the integrated terminal — not to your operating system.
That boundary is easy to underestimate until you count how much developer writing happens elsewhere:
- The pull-request review you type in a browser
- Slack, Teams or Discord threads
- The ticket description in Jira or Linear
- A second editor or IDE — JetBrains, Xcode, Android Studio, Zed
- A terminal outside VS Code: iTerm2, Windows Terminal, a tmux session over SSH
- The incident post-mortem in Notion or Confluence
- Commit messages composed in a Git client
None of those can receive dictated text from VS Code. That is not a defect, it is the design. The roadmap issue in the vscode repository — “Extend built-in on-device dictation to editor dictation, terminal voice, and add read-aloud (TTS)” — is about widening coverage within VS Code, and it explicitly plans to “fall back to the extension where built-in on-device transcription is unsupported”.
A system-wide dictation app inverts the model: one global hotkey, transcription in the background, text inserted at the cursor of whatever window has focus. The receiving application never needs to know the tool exists, which is why the same setup covers a Jira field and voice dictation for code documentation without any per-app integration.
VS Code dictation vs Weesper Neon Flow: what each one covers
The two tools solve different halves of the problem. One is deeply integrated into a single window; the other is shallow but universal.
| Capability | VS Code built-in dictation (1.131) | Weesper Neon Flow |
|---|---|---|
| Where the text lands | Chat inputs, text editors, integrated terminal | Any macOS or Windows app, at the cursor |
| Audio processing | On device; model downloads on first use | On device, local Whisper |
| Transcript sent off-device? | Only with dictation.experimental.llmCleanup on | No |
| Maturity | Experimental | Shipping product |
| Operating systems | Windows x64/Arm64, macOS Apple silicon, Linux x64/Arm64 (glibc 2.34+) | macOS, Windows |
| Explicitly unsupported | VS Code for the Web, Intel Macs, 32-bit/Arm32, musl Linux | Linux |
| Languages | 40 language-locales in the default model’s own card; editor list not published | 55+ |
| Trigger | ⌥⌘V editor, ⌘I chat, Command Palette for terminal | One global hotkey, in every application |
| Cost | Included with VS Code | €5/mo, €45/yr or €99 one-time; 15 days free |
The language row is the one most often misread. The 40 language-locales figure belongs to the NVIDIA model, measured by NVIDIA in 2026; VS Code has not published which of those it surfaces.
Which one should you choose as a developer?
If your dictation never leaves VS Code and your machine is on the supported list, the built-in feature is the obvious default. If your writing spans a browser, a chat client and a second editor, a hotkey that works everywhere is worth more than a deeper integration in one window.
You work almost entirely inside VS Code. Built-in dictation wins on friction alone: nothing to install, nothing running in the background, one microphone session shared by chat, editor and terminal.
Your day is spread across a dozen windows. The built-in feature covers one of them. Everything else — the PR comment, the standup note, the customer email — still gets typed. This is where a system-wide dictation tool tells you more than any feature table, because the difference only shows up in the applications VS Code cannot reach.
Your hardware is not on the list. An Intel Mac, a browser-based workspace or an Alpine container gets nothing from built-in dictation. A system-wide app on macOS or Windows sidesteps the question entirely, though it will not help you on Linux, where Weesper does not run.
Your transcripts are covered by an NDA. Leaving dictation.experimental.llmCleanup off keeps everything local in VS Code. If the requirement is that no transcript ever leaves the machine under any setting, a tool with no cloud path at all removes the configuration risk.
Whichever you pick, general-purpose speech models rarely spell internal identifiers correctly on the first pass. kubectl, useEffect and a class name invented last sprint are exactly what a custom vocabulary for framework and library names exists to fix.
Can you run both without a hotkey conflict?
Yes. They use different triggers, and VS Code’s dictation can be switched off entirely with dictation.enabled.
- Decide which engine owns the editor. For one consistent behaviour everywhere, set
dictation.enabledtofalseand let your global hotkey handle VS Code like any other window. - If you keep both, avoid the collision. VS Code claims
⌥⌘V/Ctrl+Alt+Vfor editor dictation and⌘I/Ctrl+Ifor chat. Choose a global hotkey that touches neither. - Keep LLM cleanup off on repositories you cannot share. It is a per-setting decision, not a per-project one, so it is easy to forget.
- Test the terminal case specifically. VS Code’s terminal dictation runs from the Command Palette; a global hotkey works in any terminal, integrated or not.
Running both is a reasonable end state: built-in dictation for quick chat prompts, a system-wide tool for everything else. That combination is what makes a voice-driven coding workflow survive contact with a real working day, rather than only with a demo.
The short version
VS Code 1.131 gave developers something worth having: private, on-device dictation in the editor, terminal and chat, with no extension and no cloud round-trip for the audio. Within its boundary it is excellent, and it is experimental, so expect the settings to move.
The boundary is the whole story. Built-in dictation covers one window. A dedicated app covers the operating system — which is where pull-request reviews, ticket descriptions, chat replies and second editors actually live.
Next step: try Weesper Neon Flow free for 15 days and dictate into the applications VS Code cannot reach, or check the hotkey and microphone guides in the Help Center before you set your global shortcut.