The hard case isn't a second language. It's one sentence with both.

Most language-support marketing answers a simpler question than the one a genuinely bilingual writer actually has. "Does it support Spanish" is easy — plenty of tools do. The harder, more common case for someone who actually works across two languages is a single sentence that starts in one and finishes in the other, or a name and a technical term that don't translate and shouldn't be forced to. This page checks five tools against both questions: the published language count, and what happens to a sentence that doesn't stay in one language for its whole length.

The alternatives

Wispr Flow — the widest published language list

Checked on wisprflow.ai/pricing today, the page states: "Speech-to-text across all your apps in 100+ languages." Pro is $15 a month (20% off billed annually), on Mac, Windows, iOS and Android, with a vocabulary that learns names and jargon and syncs across devices — useful for anyone who dictates the same proper nouns in more than one language. Their site doesn't make a specific claim about mixing two languages inside one sentence; the 100+ figure describes breadth of coverage, not that particular behaviour. The comparison covers the rest.

superwhisper — the same breadth, plus translation

Checked on superwhisper.com today, the site states: "Superwhisper supports 100+ languages & dialects. It can also translate them all to English." That translation step is a distinct feature from transcription accuracy in the spoken language — useful if the end goal is an English document regardless of what language you're thinking in, less useful if you specifically want the output to stay in the language you spoke. Pricing, checked on their comparison page, is $8.49 a month, $84.99 a year, or $249.99 once. The comparison has the rest.

MacWhisper — 100+ languages, and a file-transcription path too

Checked on macwhisper.com today, the site states it "Supports over 100 languages for seamless multilingual transcription." Free is $0 forever for core transcription; Pro is a €64 one-time purchase adding system-wide dictation and grammar cleanup. Its underlying engine is also used for transcribing existing audio and video files, so a recording made in one language, or a meeting that drifted between two, can be processed after the fact rather than only live. The comparison covers the rest.

Apple Dictation — free, with a language list that depends on your Mac

Checked against Apple's own dictation guide today: the page states "Dictation is not available in all languages or regions, and features may vary," and points to a separate macOS feature- availability page for the current list rather than publishing a fixed one directly. You can add multiple languages in Keyboard settings, and once more than one is configured, you can switch between them mid-dictation by clicking the language abbreviation that appears near the cursor — a real, if manual, way to move between languages inside one working session. The comparison has the rest.

Google Voice Typing — the longest list, with an English catch

Google's own page for voice typing inside Docs lists well over a hundred languages for transcription and is free with no separate charge. Its editing-by-command layer — "select paragraph," "bold," "insert table" — only works in English, and Google's own help states the account language and the document language both need to be English for those commands specifically. The transcription breadth is real; the fuller feature set narrows sharply outside English, and it only works inside Google's own web apps rather than typing into any Mac application. The comparison has the rest.

Coii VoiceInput — a smaller list, and one specific trick none of the others state

$19 once, three devices, macOS 13 or later. The default engine detects the spoken language itself and handles Chinese and English mixed inside a single sentence, applying punctuation to the Chinese half without a separate step. A second, larger engine covers a broader set — described only as "dozens" of languages rather than a specific count — plus the option to bring your own model. Languages can also be set explicitly, among Chinese, English, Spanish, French, German, Japanese, Korean, Russian, Portuguese and Italian, or left on automatic detection. This is a narrower published list than Wispr Flow's, superwhisper's or MacWhisper's 100+, and that gap is worth stating plainly: if your requirement is breadth across many named languages, the other three publish a larger number.

The table

The gap worth naming plainly

Three vendors on this page publish a triple-digit language count and this app does not — it ships a smaller named list plus a broader engine described only in general terms. If the actual requirement is a specific language outside that named list, Wispr Flow, superwhisper or MacWhisper is very likely the correct answer, and no amount of framing changes that. What this app does instead is narrower and more specific: real-time code-switching inside a single utterance for one language pair, rather than broad single- language coverage across many. Those are different kinds of "multilingual," and which one matters depends entirely on whether your daily writing lives inside one language at a time or moves between two mid-sentence.

What actually goes wrong across languages, regardless of the app

Accented and non-English names remain the hardest case for every engine on this page, on-device or cloud. None of these vendors, including this one, publish an accuracy figure broken out by language, and a name the model has never encountered is likely to come out wrong the first time regardless of which tool transcribed it, in any language. Reading back anything with unusual proper nouns before it's sent is the honest advice here no matter which app did the listening.

What a mixed-language working day actually looks like

For someone who genuinely thinks and writes in two languages depending on who they're talking to, a working day rarely separates cleanly into "the Spanish hour" and "the English hour." A Slack message to one colleague, an email to another, a note to yourself that starts in one language and drifts into the other mid-thought because that's simply how the thought arrived — that's the actual pattern, and it's a different requirement from "I occasionally need to dictate in Spanish." Tools built around a single selected language, switched manually when needed, handle the first case adequately and the second one with real friction: every switch is a stop, a menu, and a restart of the sentence. A tool built to detect the language per utterance removes that stop entirely, at the cost of only working for the specific language pair it was built to detect.

Regional accents inside one language

Language and accent are separate variables that a published country count can quietly obscure. A tool listing "Spanish" as supported doesn't distinguish between Castilian, Mexican, and Rioplatense pronunciation unless it says so explicitly, and none of the vendors on this page break their published numbers down by regional accent. The practical fix is the same one that applies to unusual proper nouns generally: test with your own voice, in your own accent, on your own real sentences, rather than trusting a published country count to predict accuracy for your specific speech.

Automatic detection versus a manual switch

The difference between a tool that detects language automatically and one that requires selecting it first is easy to underestimate until you're mid-sentence. A manual switch, like Apple Dictation's language abbreviation next to the cursor, works well when you know in advance which language a whole message will be in — draft an email, pick the language, dictate it. It works less well the moment a single thought needs both, because there's no way to switch languages inside one utterance without stopping to click first. Automatic per-utterance detection removes that stop but narrows the practical language pairs on offer to whichever ones the vendor built and tested detection for — breadth and fluidity are, in this specific respect, a genuine trade-off rather than both being fully available at once.

What "supported" doesn't promise

A language appearing on a published list is a statement that the vendor has built and shipped recognition for it, not a promise of parity with their best-supported language. Accuracy on a widely-spoken language with a large volume of training material available tends to be meaningfully better than accuracy on a less common one, even inside the same published "100+" figure, and none of the vendors here break their number down by expected accuracy per language. If a specific language is unusual or under-resourced, the published count is a starting point for testing, not a guarantee of the result — the only way to know is to dictate your own real sentences in it and read back what came out, the same advice that applies to unusual proper nouns and regional accents throughout this page.

What switching costs, specifically for a multilingual writer

Moving between these tools costs more for a multilingual writer than for a single-language one, because the thing most likely to be lost isn't a document format — every tool here types plain text — it's whatever per-language tuning was built up. A learned vocabulary in Wispr Flow or superwhisper that has adapted to names and terms in two languages resets on switching away. A manually configured language list in Apple Dictation is a five-minute rebuild at most. Coii VoiceInput's automatic detection needs no rebuilding at all, since there's no per-language profile to lose in the first place — the trade-off for that is the narrower published language list noted above.

The keyboard-language trap

One failure mode is common enough across this category to name directly: a dictation tool typing correctly-recognised text into the wrong keyboard layout or autocorrect language, which then mangles it a second time after the dictation itself succeeded. This isn't unique to any one vendor and none of them, including this one, fully control what happens after text lands in a target application — a text field set to expect one language will sometimes autocorrect a perfectly accurate dictation in a different one. Checking the target app's own language setting, not just the dictation tool's, is worth doing when a mixed-language sentence comes out looking wrong, since the dictation may not be where the error was introduced.

Testing before committing to any of them

Every vendor on this page offers some way to try before paying — a free tier, a trial, or a purchase that comes with a refund window. For a multilingual writer specifically, the useful test isn't a generic sample sentence from a vendor's own demo; it's a paragraph of your own actual writing, in the actual language mix you use day to day, read at your normal speaking pace. That single test surfaces the accent handling, the mid-sentence switching behaviour and the punctuation quality that a published language count can't tell you on its own, and it costs nothing more than the few minutes it takes to dictate and read back.

Picking one

If your requirement is breadth — a specific one of a hundred-plus named languages — Wispr Flow, superwhisper or MacWhisper each publish that number and this app does not try to compete with it. If your actual daily writing mixes Chinese and English inside the same sentences and switching a setting between them is the friction you're trying to remove, the trial covers thirty days of the one thing this app is built specifically to do.