Originally published at devtoolpicks.com
You think faster than you type. That's the whole argument for voice coding tools, and it holds up better than most productivity claims because the numbers aren't close. Speaking runs at roughly 150 words per minute. Typing sits somewhere between 40 and 70 for most developers.
That gap didn't matter much when you were writing code by hand. It matters a lot now that half your day is spent explaining things to an agent in full sentences.
Here's the part nobody selling these tools will tell you: if you work mostly inside Claude Code, the built-in /voice command is free, costs zero tokens, and is probably all you need. Everything else on this list earns its money by working in the other twenty apps you use.
Quick verdict
| Tool | Best for | Price | Rating |
|---|---|---|---|
| Claude Code /voice | Anyone already on a Claude plan | Free | 4.5/5 |
| Wispr Flow | Dictating everywhere, on every OS | $15/mo, or $12/mo annually | 4/5 |
| Superwhisper | Power users who want per app modes | Free tier, $849 lifetime | 3.5/5 |
| VoiceInk | Local dictation without a subscription | $29 lifetime, 1 Mac | 4/5 |
| SKI | Hearing your agent talk back | Free | 4/5 |
| Heard | Narrating several agents at once | Free tier, self host free | 3.5/5 |
Why did this category suddenly split in two?
Until recently, "voice coding" just meant dictation. You held a key, spoke, and your words landed in the prompt box. Every tool competed on accuracy, latency, and whether it understood the word nginx.
A second category has now appeared: tools that speak the agent's reply back to you. The reason is parallelism. Running one agent, you sit and watch the terminal. Running four, you can't. Reading became the bottleneck, not typing. So tools like SKI and Heard synthesize what matters and say it out loud, which means you can walk away and still know when something broke.
Work out which half you need before you pay for anything. The pricing looks nothing alike.
Is Claude Code's built-in /voice good enough?
For a lot of people, yes. This is the honest starting point.
Run /voice in a Claude Code session and you get push-to-talk dictation. Hold space, speak, release, and the transcript lands at your cursor. There's also a tap mode (/voice tap) that starts on one press and submits on the next, which skips the brief warmup that hold mode needs to detect a held key.
The details that matter to a developer are good. Transcription is tuned for coding vocabulary, so regex, OAuth, JSON, and localhost come through correctly. Your current project name and git branch get added as recognition hints automatically, which is a small thing that saves a lot of corrections. It supports 20 languages, works in the VS Code extension and agent view, and you can rebind the key away from space if you want.
And it's free. Anthropic's docs are explicit that transcription doesn't consume Claude messages or tokens and doesn't count toward /usage limits. You're paying for the prompt, not the speech.
Who should skip it: the perimeter is narrow. It only works inside Claude Code and its VS Code extension. Dictating a Slack message, a commit message, or a GitHub issue means reaching for something else. It also needs a Claude.ai account, so it won't run if you authenticate with an API key directly or through Bedrock, Google Cloud, or Microsoft Foundry. Audio streams to Anthropic's servers rather than processing locally, and organizations with HIPAA compliance enabled have it disabled outright. No remote environments either, so SSH sessions and Claude Code on the web are out.
Try this before you buy anything. Seriously. If you spend most of your day in one agent, the paid tools are solving a problem you may not have.
What does Wispr Flow cost, and is it worth $15?
Wispr Flow is the mainstream pick, and its main advantage is coverage. Mac, Windows, iPhone, and Android, working in every text field on all of them.
The free Basic plan gives you 2,000 words per week on desktop and 1,000 on iPhone, plus a custom dictionary, 100+ languages, and a privacy mode. At normal speaking pace, 2,000 words is about thirteen minutes of talking. You'll burn through it in two or three days.
Pro is $15 per user per month, dropping to $12 if you pay annually ($144 up front). That unlocks unlimited words, command mode for editing by voice, and team features. New accounts get 14 days of Pro with no card required, and students get three months free plus 50% off after.
Pros: genuinely the broadest platform coverage in the category. The custom dictionary handles project specific terms well. Setup takes about two minutes.
Cons: everything runs in the cloud. There's no local processing option at all, which means your audio leaves your machine on every dictation. For anyone under an NDA or handling client data, that's a conversation with someone before you install it. It's also the most expensive subscription here, and at $144 a year you're paying more than the lifetime price of two of the tools on this list.
Who should skip it: anyone who needs offline dictation, and anyone whose work makes cloud audio processing a compliance problem.
Is Superwhisper still the power user pick?
Superwhisper runs on Mac, Windows, and iOS, and it's built around a modes system. You define a mode per task or per app, each with its own AI model, formatting rules, and prompt. Point one mode at Slack for casual tone and another at your editor for clean technical text.
It has real developer investment behind it. There are dedicated integration pages for Claude Code, Codex, OpenCode, and CLI use, and the app is designed to sit alongside agentic workflows rather than just being a generic dictation box.
The free tier is more useful than most. You get voice to text in any app, meeting recording and transcription, 100+ languages, unlimited use of the small local AI models, and custom prompt control. Pro adds bring-your-own API keys, unlimited cloud and large local models, translation to English, and file transcription. There's a 15 minute Pro trial, a 30 day refund window, and one license activates on as many devices as you want.
Now the part that surprised me. The lifetime license is $849. That's not a typo, and three separate comparison sites are still quoting $249.99 as of this week. Check their pricing page yourself before committing.
Pros: the most configurable tool here. Local models mean it works on a plane. The free tier is usable long term.
Cons: the price. At $849 lifetime it's the most expensive option here by a wide margin, and the break-even math against a $29 competitor is brutal. Offline models only run well on Apple Silicon, pushing Intel Macs toward cloud modes. Setup takes longer because the modes system expects configuration.
Who should skip it: anyone who just wants to press a key and talk. The configurability is the product, and if you don't want to configure things you're paying for surface area you'll never touch.
Is VoiceInk the cheapest serious option?
VoiceInk is an open source macOS dictation app from indie developer Prakash Joshi Pax. It's GPL v3 on GitHub, uses local Whisper models through whisper.cpp, and processes everything on device.
Pricing is one-time and refreshingly plain. Solo is $29 for one Mac, Personal is $49 for two, Extended is $69 for three. All include lifetime updates and a 14 day money-back guarantee. There's also a free trial, and because the source is public you can build it in Xcode and run it for nothing if you don't mind skipping automatic updates.
The standout feature is Power Mode, which applies different dictation settings automatically based on which app or URL is in front of you. Your terminal and your email client get different behavior without you thinking about it.
Pros: hard to beat on price. Auditable privacy, since you can read the code that handles your audio. No subscription to cancel and nothing to forget about.
Cons: Apple Silicon only, and it needs macOS 14.4 or later. If you have a Windows machine anywhere in your setup, this doesn't help you. It's also less polished than the commercial options, and there's no dedicated IDE integration.
Who should skip it: Windows and Linux users, Intel Mac holdouts, and anyone who wants a company with a support team behind their tools.
What does SKI do that dictation tools don't?
SKI launched on Product Hunt this week and it's the clearest example of the second category. It works with Claude Code, Cursor, Codex, Gemini CLI, and OpenClaw on both Mac and Windows.
The pitch is closing the loop. You hold a key and speak, your agent works, and then it answers out loud in a synthesized voice. Everything runs on device, both the speech recognition and the voice synthesis, so no audio is uploaded. It has full-duplex echo cancellation, which in practice means you can talk over it mid-sentence and still be heard on open speakers. There's an optional review mode that drops each transcript into an editable bubble before sending, and you can bind several repos at once with each answering in its own voice.
It's free for life according to the makers.
Pros: free, private by architecture, and it solves a problem the dictation tools don't touch. The multi-repo routing is smart if you actually run parallel agents.
Cons: it's brand new. That's the honest caveat. A tool that launched days ago has no track record on reliability, no history of shipping fixes, and no guarantee the "free for life" promise survives contact with a funding round. Voice output is also more intrusive than it sounds. A synthesized voice narrating your terminal is either a revelation or deeply annoying, and you'll know within an hour which camp you're in.
Who should skip it: anyone who works in a shared office, and anyone who needs a tool with a two year track record before it touches their workflow.
Should you use Heard instead?
Heard attacks the same problem from a different angle. It's a macOS menu bar app that connects to Claude Code, Codex CLI, and Codex App, then narrates what your agents are doing.
The interesting design choice is judgment. Rather than reading output aloud, it decides what's worth saying based on context: what tool just ran, whether something is a routine step or a decision point. When you run agents in parallel it summarizes at the project level, so you hear one coherent story instead of five terminals competing for attention.
It's open source and you can self-host with your own API keys. Bring an ElevenLabs or Speechify key for voice, or fall back to local Kokoro, plus an Anthropic key for the narration logic, or fall back to neutral templates. The managed plans suit people who'd rather download and sign in: a free tier with cloud voices, two starter personas, and roughly 10,000 characters of speech per day, then Pro at around 20,000 with all voices unlocked, then Power for hands-free control and phone pairing.
Pros: the self-host path is free if you already have API keys. Project level summarization is the right abstraction for parallel agents, and being open source means you can see exactly what it does.
Cons: self-hosting wants a Python environment and CLI comfort, which is fine for this audience but isn't a two minute install. The managed tiers cost money on top of your existing API spend. macOS only. And characters per day is an odd unit to budget against.
Who should skip it: Windows and Linux users, and anyone who wants voice input rather than output. Heard doesn't replace a dictation tool, it sits next to one.
How do you pick?
Work out which half of the category you're in first, then whether you need to leave your agent.
Superwhisper is the odd one out. It's an excellent tool at a price that became very hard to defend when the lifetime jumped to $849.
The recommendation
Start with /voice. On any Claude plan this costs nothing and burns no tokens. Use it for a week before spending a cent. A surprising number of people find their problem was already solved.
Dictating outside your agent on a Mac? Buy VoiceInk at $29. One payment, local processing, open source. Best value here and it isn't close.
On Windows or need mobile? Wispr Flow at $15/month. It's the only tool with genuinely complete platform coverage, and that coverage is what you're paying for, not better transcription.
Want your agent talking back? Try SKI first, because it's free. Just don't build a workflow around a tool that launched this week until it's proven it ships fixes.
I'd skip Superwhisper at current pricing unless the modes system solves something specific for you. If you're weighing your whole agent setup rather than just the input method, the Claude Code alternatives and Cursor alternatives comparisons cover what these plug into, and the local AI coding tools roundup pairs well with the on-device picks above.
Voice input is one of the few workflow changes that pays off immediately instead of after a week of retraining habits. Give the free option a fair trial. If you're new to the agent itself, the Claude Code walkthrough for solo developers is the place to start. Found something better? Let me know on Twitter @hafizdev.
Top comments (0)