Voice to Text on Mac Without a Subscription: Three Routes

You want to talk instead of type on a Mac, and every app you find wants a monthly fee. macOS already ships with free dictation, so the fee looks like paying twice for the same thing. It isn’t. Once you know what the money is for, you can decide whether to pay it, pay someone else, or pay nobody.

What the subscription actually pays for

A dictation app like ListenIn has two stages, and only one of them costs money per use.

Recognition turns audio into words. ListenIn hands that job to Apple’s speech engine with on-device recognition forced on, on both paths: the modern engine on macOS 26 and the classic fallback on macOS 14–15. Apple’s on-device recognizer ships with macOS and has no per-request price. Audio never leaves the Mac and never reaches ListenIn’s cleanup service. The one trade-off: on macOS 14–15, if the dictation language pack is not downloaded, recognition fails outright instead of quietly using Apple’s cloud. That case has its own post.

Cleanup is what separates a dictation app from the built-in one. The recognized text goes to a language model that removes fillers and false starts, repairs sound-alike errors, and formats, under one rule: reordered, never invented. It does not add content and does not answer the question you dictated. That model call is compute, billed per request by whoever runs the model. On ListenIn’s built-in path, ListenIn pays the provider for every cleanup, and Pro is how that gets covered.

So the recurring cost is the cleanup pass, not the speech-to-text. Everything below follows from that. The expensive part is a text-to-text call, and a text-to-text call can be routed anywhere. The security page has the mode-by-mode table of what each route sends and where.

Route 1: the free tier, no account, no card

ListenIn’s Free plan is 8,000 cleaned words per week. There is no account to create and no card to enter; the app never asks for an email. Built-in cleanup requests carry a random device ID, which is how the weekly count works without a login.

Your first cloud cleanup starts a 14-day trial with everything unlocked. When the trial ends you drop to the weekly allowance, not to nothing. When the allowance is used up, dictation keeps working: the raw Apple transcript still lands at your cursor and the HUD says “Weekly quota used up — raw transcript inserted · Upgrade to Pro”. Raw dictation has no cap. The menu shows a words-this-week counter so the cutoff is never a surprise.

If you dictate a handful of messages and notes a day, this may be all you ever use. For how this compares with other free tiers, including which ones require an account, see the comparison page; its prices and terms were checked in August 2026 and are dated on the page.

Route 2: bring your own API key

If you already pay a model provider, you can pay them for cleanup instead of paying ListenIn.

Open Settings → Models. Under Cloud (online), Add cloud model opens a sheet: pick a Provider — there are presets, plus Custom for any OpenAI-compatible endpoint — then fill Name, API Key and Model. A Base URL field appears for OpenAI-compatible endpoints. The preset list is what the app can talk to today; read it there rather than from a blog post, because it changes with releases.

What changes once you save it: text goes straight from your Mac to that provider. ListenIn’s server is not in the path, and the in-app note states the terms plainly: “Your key stays on this Mac and goes straight to the provider — ListenIn never sees it, and these requests skip the built-in quota.” No weekly cap, no trial clock. Your bill is whatever the provider charges per token, and nothing from ListenIn.

What does not change: audio still never leaves the Mac. Recognition is still Apple’s on-device engine. Each request contains your recognized text, the cleanup instructions, and, on Standard and Refined, your dictionary terms and some recent text context. That text is now under your provider’s terms. That is the trade, and it is the right trade if you already have a data agreement with that provider or want a stronger model than the built-in path gives you.

This post does not quote provider rates. They change, and you should read them from the provider before you commit.

Route 3: local Ollama, pay nobody

The zero-cloud route. Install Ollama and pull a model. In the same Settings → Models tab, the On-device (offline) section lists what Ollama has; if it isn’t installed or running, the tab offers Get Ollama or Start Ollama. Pick a model with Use, then flip Mode from Hybrid to Offline. The footer under the picker states the contract: “Offline — cleanup runs only on your on-device model. Nothing leaves this Mac; cloud is off.”

Now nobody bills you. Recognition is Apple’s on-device engine; cleanup is your own processor. The trade-offs are real:

• Setup and upkeep are yours. Ollama has to be installed, running, and holding a model.
• Cold starts. A model evicted from memory reloads on the next request, and the first dictation after that will pause.
• Quality is your model’s. There is no figure to quote for it; it depends on the model you chose and the Mac you run it on. Refined asks the most of a model. If it disappoints, drop Cleanup strength to Standard.

If Ollama is down, ListenIn does what it always does when cleanup is unavailable: the raw transcript lands, the failure is labeled, and nothing goes to the cloud instead. The setup guide with a Wi-Fi-off test covers the rest, and what “local” means in dictation apps explains why the two stages have to be judged separately.

What Pro buys

ListenIn is a subscription product and says so. Pro is $12 a month or $96 a year; the pricing section is the source if those numbers ever differ from this post.

What you get for it: the weekly cap goes away, subject to a fair-use policy spelled out on the same page; priority access during high demand; and nothing to manage. No key to paste into a sheet, no Ollama to keep running, no model to choose. You pay ListenIn to make the cleanup call and to eat the provider bill.

Buying does not create an account either. Polar handles payment and emails a license key; paste it into License Key (Pro) and tap Activate Pro. The key is the whole relationship.

What Pro does not change: the audio boundary. Pro text goes to the same built-in cleanup service Free text does, and audio still never leaves the Mac.

One-time-purchase apps with local Whisper

Some Mac dictation apps sell a one-time license and run recognition on a Whisper model you install yourself. On the comparison page, superwhisper is the example: deep model choice, any local Whisper variant, and a one-time-purchase option.

That is a different trade, not a worse one. You pick and manage the recognition model, which gives you control over a stage ListenIn does not expose; ListenIn’s recognition is Apple’s engine, full stop. In exchange, model selection, downloads, and upkeep are your job, the same way Ollama is your job on Route 3.

Whichever app you look at, apply the same test from the top of this post: which stage runs on the Mac, which stage is a per-request cloud call, and who pays for that call. A one-time price can cover software; it cannot cover unlimited cloud compute forever, so find out where the compute happens. The comparison page puts the apps side by side with dated prices, and the Typeless page covers the most common switch in detail.

Which route fits which person

You	Route
Dictate occasionally and want zero setup	Free: 8,000 cleaned words a week, no account, no card
Already pay a model provider and have terms you trust	Your own key: text goes to them, you pay them
Need nothing to leave the Mac, willing to run Ollama	Offline mode: pay nobody, quality is your model’s
Dictate all day and want nothing to manage	Pro
Want to choose the recognition model itself	A local-Whisper app; see the comparison page

None of these lock you in. There is no account to close, your dictionary and learned corrections export to a single file and import back with a merge, and switching routes is a picker in Settings → Models. The FAQ covers what this post did not. If something here disagrees with the app, that is a bug; report it.