Apple Dictation is still the easiest zero-install option. It wins on convenience, not depth.
Straight comparison for Mac dictation
Best dictation apps for Mac in 2026
If you want speech to text on a Mac, the choices narrow quickly. What matters is not the longest feature list. It is whether the app feels natural to use, stays private, gives you a clear way to fix mistakes, and fits into everyday writing without becoming its own project.
The short read.
These are the real tradeoffs at a glance: convenience, privacy, teachability, and how much of the workflow you have to assemble yourself.
Arugula is the strongest all-around fit if you want local processing and a calmer workflow for both normal writing and tools like Claude Code, Codex, or Gemini.
Whisper CLI is useful if you want raw model access, scripting, and do not mind assembling the workflow yourself.
Dragon still matters mostly to people who specifically want a legacy dictation system rather than a lighter Mac utility.
| Feature | Arugula | Apple Dictation | Dragon | Whisper (CLI) |
|---|---|---|---|---|
| Price | Free | Free (built-in) | $150+/yr | Free (open source) |
| Works offline | Yes, always | Often, depending on Apple’s mode and features | Yes | Yes |
| Privacy | 100% local, zero telemetry | Depends on Apple settings and services | Local, but phone-home license checks | 100% local |
| Saved corrections | Yes — fix once, reuse automatically | No | Yes — manual vocabulary tools | No |
| Hold-to-talk | Yes — hold key, speak, release | Toggle on/off, auto-timeout | Toggle-based | No (file-based) |
| Works in any app | Yes — cursor insertion | Yes | Yes | CLI output only |
| Coding-agent workflow | Strong fit for Claude Code, Codex, Gemini, Cursor, and terminals | Basic insertion, but no saved phrase layer or hold-to-talk flow built around prompts | Older desktop workflow, not a natural fit for modern agent tools | Powerful for scripts, but you assemble the workflow yourself |
| Memory behavior | Memory-aware; can reclaim low-risk background RAM | Handled by macOS, with no app-level cleanup layer | Heavier desktop footprint | You manage model memory yourself |
| Setup required | Install + allow permissions once | None — built in | License, account, training | Homebrew, Python, config |
| AI model | Whisper (local) | Apple Speech (hybrid) | Dragon NaturallySpeaking | Whisper (local) |
| macOS support | macOS 15+, Apple Silicon | All supported macOS versions | Legacy / limited Mac path | Any macOS with Homebrew |
| Accounts required | None | No separate app account | License / activation | None |
Where the differences actually show up.
The deciding factors are usually simple: how quickly you can start, how private the workflow is, whether correction feels built in, and whether the app behaves like a Mac utility or a separate system to manage.
Arugula vs Apple Dictation
Apple’s built-in dictation wins on convenience because it is already there. If that is enough for you, it is hard to beat. But it is still Apple’s system-level workflow: limited customization, no saved phrase memory, and less sense that the app is adapting to the names and wording you use every day.
Arugula is for the person who wants dictation to feel more personal without becoming more complicated. Hold to talk, release to type, then correct the last phrase once and keep moving.
Arugula vs Dragon NaturallySpeaking
Dragon still carries weight as the old desktop dictation standard, especially for people who needed vocabulary control. For most Mac users now, though, it is the opposite of what they want: expensive, heavier than it needs to be, and no longer especially Mac-native.
Arugula keeps the useful part of that idea and drops the ceremony. You can still teach it your words, but the interaction is lighter: correct the phrase, save it, move on. Read the full Dragon migration guide.
Arugula vs Whisper CLI
If you are technical, whisper.cpp is powerful and accurate. It is also still a tool, not a finished dictation experience. There is no system-wide hold-to-talk, no cursor insertion, and no saved correction layer waiting for you after the model finishes.
That difference matters most when you are working with coding agents. Arugula lets you talk to Claude Code, Codex, Gemini, Cursor, or a terminal prompt the same way you talk into Notes or Mail: hold a key, speak, release, keep moving. Read the full guide to using Whisper on Mac.
The bottom line.
Arugula is the clearest fit if you want private dictation on a Mac that feels calm, local, and teachable, especially if your day moves between normal writing and coding agents. It is not trying to be the biggest system. It is trying to be the one you keep using.