Best speech to text app for Mac

The short answer: what to use, and when Apple Dictation is enough

For most Mac users in 2026 the pick is a local-model dictation app, with a cloud option as the escape hatch. If you mostly send short messages, the built-in Apple Dictation costs nothing and works in any text field. The moment your dictation gets longer, more technical, or needs to land inside an app Apple does not reach cleanly, you want a dedicated tool.

The dedicated tool I would reach for runs the transcription on your machine. On an Apple Silicon Mac that is fast, private, and works on a plane with the Wi-Fi off. Whisper by Remskill runs two local engines, the open-source Whisper family and NVIDIA's Parakeet, fully on-device, no Python, no server in the loop. You hold Command and Option together to record, release either key to stop, and the text appears wherever your cursor is. If you prefer the latest OpenAI models for the highest accuracy, there is a cloud mode that uses your own OpenAI key. Same hotkey, same overlay, one toggle.

Two paragraphs, and that is the whole decision. The rest of this article is the case for it, the honest limits, and the apps you should know about before you commit.

How I picked, so you can disagree with me on purpose. I weighted four things, in order. Does the audio stay on your Mac, because privacy is not a feature you bolt on later. Does it paste at the cursor in any app, or trap your words in its own window. Does it handle long and technical speech, not just short notes. And what does it cost to use day to day. A tool can lose on raw accuracy and still win if it nails the first two. Most of them lose on the first two while shouting about the fourth. I am the kind of architect who draws the comparison table before installing a single app, and the table is always wrong by the second download. This one survived contact.

Why Apple Dictation runs out of road

A myth runs through half the Mac dictation articles you will find: Apple Dictation hard-stops after sixty seconds. That used to be true on older macOS releases, where continuous dictation was not enabled by default. Apple's current documentation says the opposite: you can dictate text of any length without a timeout, and dictation only stops on its own after thirty seconds of silence. So the sixty-second cutoff is no longer the real limit.

The real limits are quieter. Accuracy drops on technical terms, names, and accented speech, which is exactly where long-form work lives. Apple counts forty-plus language and regional variants, but it splits English (US) from English (UK) to get there, so the headline number is bigger than the distinct-language count. There is no custom vocabulary, no biasing toward your jargon, no AI cleanup of casual speech into polished writing. It is a good free microphone that types. It is not a writing tool.

That gap is the entire reason "Apple Dictation alternative" is a search people make. They tried the free one. It got them to short notes and stopped being enough.

Local vs cloud on a Mac: privacy, speed, and the accuracy gap

Every Mac dictation app sits on one side of a line: it runs the model on your machine, or it sends your audio to a server. The line matters more than the marketing does.

Local-first means the audio never leaves your Mac. After a one-time model download, transcription works offline: no internet, no account check, nothing in a vendor's logs. For the email to your kid's school, the salary spreadsheet, the brief you are drafting, that is the right default. On Apple Silicon it is also fast, because the model runs on hardware that is sitting idle while you talk. The first time you dictate a paragraph on a plane with the Wi-Fi off and watch the text appear, the cloud-versus-local debate stops being abstract.

One accuracy lever almost nobody mentions beats the model choice: your microphone. A decent USB mic does more for transcription accuracy than jumping from a small model to a large one. If your words come out garbled, fix the input before you blame the engine. I spent a full evening swapping models to fix what turned out to be my laptop sitting next to a running dishwasher. The boring fix is usually the right one.

Cloud means the latest models and the best raw accuracy. OpenAI's transcription models lead real-world accuracy, which is why a cloud option earns its place for the hardest audio. In Whisper's cloud mode you bring your own OpenAI key. Transcription runs on gpt-4o-mini-transcribe or gpt-4o-transcribe, and the audio goes straight from your machine to OpenAI. We are never in the middle, and we store no audio or transcripts in either mode.

I have watched cloud-everything go wrong. A team I worked with had a contractor build an internal dictation prototype on a frontier model, calling the API for every utterance, with retry logic that was a little too eager. The cloud-cost dashboard at quarter-end showed a five-figure bill, most of it from transcribing the same standup recordings four times. The CFO's take was simple: maybe we should not pay a server to transcribe meetings that already have notes. On a Mac, you usually do not need the server. Cloud is the escape hatch, not the default.

The Mac dictation apps worth knowing

You did not come here for one option, so here are the ones that show up when people compare Mac dictation tools (yes, including the two named after the same open-source model, which keeps the support inbox lively). One-line takes, honest where the gaps are.

• Apple Dictation: free, built in, works in any text field. Best for short notes; no custom vocabulary and accuracy fades on long or technical speech.
• Whisper by Remskill: two local engines plus a BYOK cloud mode, push-to-talk hotkey, pastes at the cursor in any app, free for the whole local pipeline.
• Superwhisper: a Mac Whisper-based app that works offline, advertises 100-plus languages with translate-to-English, and has a free tier after a short Pro trial. If you are weighing the trial wall and the word cap, here is our free superwhisper alternative.
• MacWhisper: a popular Mac-only wrapper that runs Whisper models locally, with a free version and a paid Pro, sold through Gumroad.
• Wispr Flow: a cloud voice-to-text app across Mac, Windows, iPhone, and Android, advertising 100-plus languages with auto-detect and AI auto-editing of casual speech. Cloud-based, not offline-first.
• Dragon (Nuance): the old heavyweight, but Nuance discontinued the individual Dragon product for Mac years ago, so it is not a live option for new Mac users.
• Otter.ai: a meeting-notes tool that joins Zoom, Teams, and Google Meet; a different category from cursor dictation, with a free tier capped at 300 monthly minutes.

Here is a real-specs comparison, not a wall of "fast" and "easy."

A note on those language counts, because most roundups get them wrong. Whisper's 99 languages applies to the multilingual models only; the English-only builds do English and nothing else. Parakeet covers English plus 24 European languages, twenty-five in total, and no Asian languages or translate-to-English. OpenAI's cloud transcription lists around fifty-seven languages. "Supports every language" is a marketing sentence, not a spec.

Picking the right setup for your Mac

The right pick depends on two things: your Mac, and what you dictate. The app does not choose for you. It presents three paths and you pick.

If you have an Apple Silicon Mac, any M-series chip, you have the whole menu. Local Parakeet is the fastest, running five-to-ten times faster than Whisper on CPU, and it is plenty for everyday English dictation. Local Whisper is slower but supports 99 languages, translate-to-English, and custom vocabulary, so it is the pick if you switch languages or need control. Cloud mode is the pick when you want the latest OpenAI accuracy and do not mind paying OpenAI directly through your own key.

Model size is the other lever. The local Whisper lineup runs from a roughly 140 MB Base model up to a 3 GB Large v3, with Parakeet sitting around 600 MB. On a Mac with 8 GB of RAM, stick to Parakeet or the smaller Whisper models and skip the 1.5 GB-plus ones. On a 16 GB Apple Silicon Mac, everything runs. If you are on an older Intel Mac, cloud mode runs on any hardware, and the small local models will work where the large ones choke, though Apple Silicon is where the desktop app is built to shine.

To set the hotkey: on macOS the default is holding Command and Option together as push-to-talk, and you can rebind it in Settings to anything that does not collide with Spotlight or the system dictation key. If you want the full walkthrough, our voice-to-text on Mac guide covers the setup step by step, and the offline speech-to-text guide goes deeper on running everything with the network off.

When to skip our app and use something else

I will say the quiet part. If all you do is fire off thirty-word texts and the odd short note, do not install anything. Apple Dictation is free, built into your Mac, and works in any text field. We start being worth the download somewhere past the point where Apple's accuracy and lack of custom vocabulary start to hurt: long-form writing, technical jargon, multilingual work, dictating inside apps where you want clean paste-at-cursor behavior. If you mainly need to capture multi-speaker meetings with summaries and calendar links, that is Otter's job, not ours; its free tier gives you 300 minutes a month. Different category. Use the right one.

Pricing

Whisper is free for every authenticated user across the entire local pipeline: both local engines, all the languages, AI enhancement through Ollama, history, presets, custom hotkey, hardware acceleration. No payment method at signup. The thing you pay for is Whisper Pro, which adds the cloud surface: OpenAI cloud transcription, cloud AI enhancement, and voice web search through your own OpenAI key. The plan options live on the pricing page so the numbers are always current. Apple Dictation, for the record, stays free. That is the floor every paid Mac dictation tool has to beat.

Two days after that kitchen-table demo, my daughter came back with a complaint: the hotkey did not work in her drawing app. She did not know what a hotkey conflict was. She just knew it stopped working. That is the whole job of a good Mac dictation app: not to win the feature-count war, but to disappear into the act of writing, and to be fixable when it does not. The best one for you is the one you forget you are using.