The other one is free. Read this before you spend $19.

There is no way to write this page that does not begin with the awkward part, so: Murmur — the app at murmurtype.com, which its own site calls "Free Offline Voice Dictation for Mac" — costs nothing. Not a free tier with a word cap, not a trial. Its site says "Free forever" and "No subscription, no cloud, no limits", and the source is on GitHub under the MIT licence, which means you can read every line of it, build it yourself and keep using it regardless of what anyone decides later.

It is also, on the evidence of its own description, the same shape of tool as this one: hold a hotkey, speak, release, and the text goes in at the cursor, entirely on the machine. So the question this page has to answer is not which is better in the abstract. It is what $19 buys that zero does not, and whether that is worth it to you specifically.

What Murmur says it does

Taken from murmurtype.com and its repository on 4 September 2026:

• Push-to-talk with a customisable hotkey, working "in any app" — browsers, editors, terminals, email.
• "100% on-device" voice data, and "Fully offline — no account needed."
• macOS 14 Sonoma or later, on "Apple Silicon or Intel Mac", with microphone and accessibility permissions granted.
• A local history you can edit, and learning from your corrections over time.
• An optional cleanup step that connects to "any OpenAI-compatible API" to remove filler words and tidy the text — configured by you, not shipped on.
• Recognition covering Chinese, English, Japanese, Korean and Cantonese, with punctuation handled.
• MIT licensed, source public.

Two of those are things this app does not have at all. It does not learn from your corrections, and it does not build a personal vocabulary of any kind — so the surnames it gets wrong in week one it will still get wrong in week ten unless you fix them by hand. And if you write in Cantonese or Korean, Murmur names those languages explicitly while this app publishes no comparable list; where a vendor states something and we do not, the row goes to them.

Who should stop reading and install the free one

Genuinely, and first, because burying this would be dishonest:

• You dictate occasionally. A few times a day, short things, into apps that behave. The differences below are about volume, and at low volume they do not exist.
• You want to read the code. MIT, public, buildable. Nothing on this side matches that and nothing on this side pretends to.
• You write in Cantonese, Korean or Japanese and that is the point. They name those languages; we do not.
• Money is the constraint. Zero beats nineteen and no amount of writing changes that.
• You want to learn how these things work. A readable repository is a better teacher than a licence key.

None of that is a hedge. A free, offline, open-source dictation tool that works is a good thing to exist, and somebody who installs it and never thinks about this category again has made a perfectly sound decision.

The four things $19 buys

Everything below is a behaviour, not a feature bullet, and each of them only becomes visible when you are dictating enough that the app disappears into the background.

The words are saved before the engine is asked anything

This is the one that is architectural rather than incremental. When you release the key, the transcript is committed to a local history first, before anything downstream is consulted. Only then does the rest of the sequence run.

Murmur's site describes a local history you can edit, which is a good feature and a normal one. What it does not describe is when the row is written relative to the recognition finishing, and that ordering is the entire claim here. In a design where the row appears after everything else succeeds, a failure anywhere in the chain is a lost sentence. In this one the worst case is opening the history and copying a row you already have.

If you have never lost a dictation you will find this abstract. If you have, you will not.

The target is decided when the key goes down

Press the key while your cursor is in a Slack message box, talk for eleven seconds, and in that time a notification arrives, a build finishes, or you tab away to check a name. The text still lands in the Slack box. The destination was fixed at key-down and is not re-read when you let go.

This sounds like a corner case until you count how often you look away mid-sentence. It is the single most common way dictation goes wrong in an ordinary working day, and it goes wrong silently — the words appear somewhere, just not where you were.

Your clipboard comes back

Where an application does not expose its text field through the accessibility route, the text is delivered by a synthetic paste — and what was on your clipboard beforehand is put back afterwards. Copy a URL, dictate a sentence, paste the URL: it is still the URL.

There is a person on the other end

This is the awkward one to argue and the honest reason a lot of paid software exists. $19 buys a notarised build, updates included with no second purchase, three devices on one licence, and an address that answers when something is broken on your particular Mac. A public repository owes you nothing on any of those, which is not a criticism of it — it is the deal, and the deal is a good one at the price.

The table, since you will want one

Count them honestly: seven rows to Murmur, four to this one, and the rest a draw. If you are choosing on row count, the free one wins and you should take it. Rows are not weighted, though, and three of the four on the left are about what happens when something goes wrong, which is a category a table is bad at representing because it happens on Thursday rather than during an evaluation.

Where they are genuinely the same

It is worth being clear about how much of this comparison is a draw, because comparisons in this category usually turn on "does your audio leave the machine" and this one cannot.

Neither app uploads your voice. Neither needs an account. Both work on a plane. Both put text at the cursor in whatever you are using. Both are push-to-talk rather than always-listening, which is the correct design and not a universal one. Both ask for the microphone and accessibility permissions macOS requires, and both will look broken until you grant them.

Against a hosted service, the privacy argument does all the work. Here it does none of it, and what is left is behaviour under failure, which is why this page is shaped the way it is.

The Ventura line

One row in the table is a hard fork rather than a preference. Murmur asks for macOS 14 Sonoma; this app asks for macOS 13 Ventura. If your Mac is on Ventura and staying there — a work machine on a managed update schedule, an Intel machine you would rather not push forward, a laptop that runs one piece of software that broke last time — then there is no decision to make, because only one of these two installs.

That is a smaller population every year and it is not zero, and it is the only row on this page where the free option is unavailable rather than merely different.

The word "free", read carefully

Free is not one thing, and the distinction matters more here than in most categories because it decides what happens in two years rather than what happens today.

Murmur is free in both of the senses people mean: no money, and a licence that lets you read, modify and redistribute the code. That second one is the durable half. A project can stop being updated, change direction, or be maintained by somebody else entirely, and an MIT-licensed copy on your disk keeps working either way. Nobody can take it back. That is a genuinely stronger guarantee than any commercial promise, and it should be weighed rather than waved at.

What free does not include is anybody's obligation. If a macOS update changes how text insertion behaves in some application you depend on, a paid product has a reason to care and a public repository has a maintainer with a day job. Both of those situations resolve; they resolve on different timescales, and which one you want depends entirely on whether dictation is a convenience or the way you get your work done.

The version of this comparison that is dishonest goes further and implies that free software is risky. It is not, and this page is not going to say so. The sentence that is actually true is narrower: the money buys a claim on somebody's attention, and if you do not need that claim you should not pay for it.

Sixty presses, not six

Every argument on this page scales with volume, so it is worth being concrete about what volume looks like.

Somebody who has genuinely switched to dictating uses it perhaps sixty times a day, and most of those are under five seconds: a two-line reply in Mail, a sentence into a Slack thread, a commit message, a comment on a document, a note to themselves. It is not the long-form writing session people imagine when they hear "dictation software". It is the constant small typing that makes up most of a day at a keyboard.

At six presses a day, nothing on the left column of the table has ever happened to you. At sixty, all of it has happened by the end of the month: you will have looked away mid-sentence, you will have had a paste land in an application that handled it oddly, and you will have copied something important and then dictated on top of it. Each of those costs somewhere between two seconds and a paragraph, and the reason to care about them is not the individual cost but the fact that they teach you to hesitate. An app you hesitate before using is an app you use six times a day, and at six times a day it has changed nothing.

That is the whole of the argument for spending money here, and it is a volume argument rather than a quality one. Both apps recognise speech well. Only one of them has been designed around the assumption that you will be doing this all day and that things will go wrong while you do.

The neighbours, briefly

This bracket has more than two apps in it, and the free-versus-paid axis is not the only one.

If what attracts you about Murmur is the open repository rather than the price, VoiceInk is also open source and is sold as a one-time purchase, with a custom dictionary and modes this app does not have. If you want a larger local application that transcribes files as well as dictating, superwhisper and MacWhisper both do that and neither of them is free. And if the reason you are looking at all is that a subscription renewed and annoyed you, Wispr Flow is the page that puts the arithmetic side by side.

Testing both properly

Do not evaluate these by dictating a paragraph and grading the transcript. Both will do fine, and you will learn nothing that decides anything.

Install one, set the hotkey, and use it for everything for a week — including the things you would normally type, and especially the short ones. Then install the other and do the same. Two rules make the test valid: change one of the hotkeys, because two apps listening for the same key is the one configuration that reliably breaks and it will look like a bug in whichever you blame; and keep a note of every time the text went somewhere unexpected, arrived late, or did not arrive. That list is the comparison. Everything else is a feeling.

At the end you will have a number for each app: how many times a day you reached for it. That number, not accuracy, is what determines whether dictation changes how much you type.

What switching costs

Almost nothing in either direction, which is unusual and worth stating. Dictation apps hold no library you would be trapped by. There is no document format, no project file and no exported archive to worry about; the output has always been text and it went into your own documents at the time.

Moving to this one means installing it, granting two permissions, setting a hotkey and re-learning nothing. Moving away means dragging it to the bin. The history is a local list on the machine you dictated on, and if you want to keep what is in it, copy the rows out before you go — same in either direction.

The thirty-day trial is deliberately longer than the decision needs, because the useful test is a fortnight of ordinary work rather than an afternoon of evaluation.

Which one to install

Take Murmur if you want the source, you dictate lightly, you write in the languages it names, you are on Sonoma or later, or you simply object to paying for something that exists for free. All of those are good reasons and none of them is a mistake.

Buy this one if you dictate constantly, you have lost a sentence before and did not enjoy it, you are on Ventura, you want the clipboard and the target window to behave without thinking about it, or you want an address to write to when something is wrong. One price, one tier, and the trial is the whole app for thirty days with no card, which means you can hold both open and let a fortnight decide.

For the neighbouring decisions: VoiceInk is the other open-source Mac option and is sold rather than free, BetterDictation is the other one-time purchase in this bracket, and if you have not yet tried what your Mac already ships with, Apple Dictation costs nothing either.