You know what you want to write.
Your input method should too.

01 · Desktop platforms

ATOK

Current retail ATOK Passport covers Windows, macOS, Android, and iOS. JustSystems previously sold ATOK X3 for Linux (discontinued; compatibility notes remain on atok.com). Separate embedded ATOK builds exist for OEM/Linux devices but are not the desktop product most users install.

02 · Shin / kyū kanji toggle

ATOK · Partial

ATOK has no global mode toggle like marinaMoji (Ctrl+Shift+3 / Ctrl+Shift+F / toolbar 新↔舊): everyday conversion still targets modern shinjitai. Traditional forms are available case by case through the character palette (文字パレット): look up variants (異体字, e.g. 国 / 國 / 圀), use radical/stroke search, or open variant lists from a committed character. Some ATOK builds also offer a conversion option that surfaces old glyphs when you type hint phrases (user reports give examples such as きゅうじたいのさい → 齋). Name and place dictionary entries occasionally include kyū forms in the candidate list. That helps for fixed characters and proper nouns, but it is not document-wide kyūjitai input comparable to a Chinese-style traditional/simplified switch.

03 · Historical kana input

Others · Partial

Stock IMEs are not useless here — Microsoft IME, Mozc, Kotoeri, and ATOK all let you reach obsolete kana such as ゐ and ゑ (for example via wyi / wye, or by converting い / え and picking from candidates). We mark them Partial because coverage is basic, conversion data stays modern-biased, and there is no systematic rekishiteki kanazukai workflow (e.g. 今日 as けふ, not きょう). marinaMoji is Full on this row because it extends the romaji table further (e.g. vu, xwa, ye combinations) and documents the limits openly; a global orthography toggle and richer dictionary conversion remain on our roadmap below.

04 · Manyōshū mode

Others · No

In standard katakana input (カタカナ入力), the IME composes in katakana: you see カタカナ in the preedit, and kanji conversion is limited or awkward compared with hiragana mode. marinaMoji's Manyōshū mode is different: the engine still converts like hiragana (full dictionary lookup, kanji + kana), but preedit, candidates, and commits are shown in katakana — the workflow scholars want for katakana-kan majiribun 仮名交じり文 and premodern all-katakana input. On marinaMoji, the toolbar "katakana" entry uses this mode. Stock Mozc, Microsoft IME, Kotoeri, and ATOK do not offer this combination.

05 · Odoriji palette & default

Others · No

Other IMEs still know 々 in fixed vocabulary (人々, 佐々) and may offer a few repetition marks if you type odoriji / おどりじ or onaji / おなじ and pick from conversion candidates. That is not the same as marinaMoji's palette (eight historical marks: 々, ゝ, ゞ, ヽ, ヾ, 〻, 〱, 〲), a session default reused with one shortcut, and toolbar integration. The following row covers applying that default to dictionary output (e.g. 人々 → 人〻).

06 · Kaeriten page layout

ATOK · Partial

ATOK does not insert kaeriten from the IME the way marinaMoji does (;r, ;1, symbols palette, Unicode Kanbun block). JustSystems' word processor Ichitaro (一太郎) — often sold alongside ATOK — is where their kanbun workflow lives: the 文字パレット「漢文」 places kaeriten 返り点, okurigana 送り仮名, and related marks through a palette-driven "apply to this kanji" workflow (JustSystems FAQ). That is analogous to our word-processor plugins (layout in the app you write in), not marinaMoji's IME-first Unicode source. Integrated proprietary stack on Windows / macOS only — not an open plugin for LibreOffice, Word, OnlyOffice, or Linux. Mouse-heavy compared with marinaMoji's keyboard-first marks + optional render step.

07 · Usage data sent to vendor

Mozc · Opt-in

Mozc's source tree includes a "send usage statistics" setting that Google documents as opt-in (off unless you enable it). On official Google Japanese Input builds — mainly Mac (checkbox in preferences; default off) and Android — turning it on can send usage data to Google. Open-source desktop Mozc (e.g. ibus-mozc, typical community macOS packages) is built without Google's upload channel: on Linux the control is hidden; in self-built/OSS binaries the backend is a no-op, so nothing is sent in practice even though the option still exists in upstream code. marinaMoji keeps that control visible for transparency but permanently disabled — Impossible for our builds.

08 · Usage data sent to vendor

Kotoeri · Partial

We use "Kotoeri" here for Apple's built-in macOS Japanese IME (the old Kotoeri name; replaced in OS X Yosemite but still widely used). Apple does not offer an IME setting comparable to Microsoft's "save/send usage data." Conversion learning and candidate history are described as local on your Mac; Private Mode pauses that learning. We cannot audit closed-source macOS line-by-line, so we do not claim Impossible. What can reach Apple: user dictionary entries sync across devices when you use the same Apple Account and iCloud; and, separately, Share Mac Analytics and related Analytics & Improvements toggles (OS-wide, opt-in/out at setup — Apple states reports avoid identifiable personal content and may use differential privacy). Improve Siri & Dictation targets voice/dictation samples, not ordinary kana→kanji typing.

09 · Usage data sent to vendor

Microsoft · ATOK

This row is about conversion habits, keystrokes, or learning data leaving your machine for the IME maker — not local learning on disk, and not marinaMoji's encrypted sync to your own cloud folder. Microsoft IME: Opt-out — Windows can collect "inking and typing" / usage data when diagnostic settings allow it; IME settings may offer "save/send usage data" but are often tied to Windows privacy toggles. ATOK: Opt-in for vendor cloud — JustSystems states that core ATOK cloud services do not collect your typed words for advertising, but ATOK Passport premium adds opt-in cloud features (Sync, cloud dictionaries, ATOK MiRA generative rewrite, etc.) that process text on JustSystems infrastructure when you use them.

10 · Active keyboard layout

Microsoft · ATOK · Partial

On macOS and Linux, the IME stack is tied to the keyboard layout you already use (French AZERTY, US Dvorak, Bépo, etc.): Latin and romaji follow the characters your layout produces, not a hidden QWERTY assumption. Kotoeri and Mozc behave this way; marinaMoji inherits it and adds physical-key shortcuts on macOS (Ctrl+Shift+1–5 stay on the number row even when the layout label is Dvorak or AZERTY) plus layout-aware macron input. Microsoft IME on Windows is Partial: direct/Latin input can follow the OS layout, but kana and romaji modes often still feel QWERTY-centric; Dvorak users frequently need the legacy IME and registry workarounds. ATOK is Partial: on Mac you can set "keyboard layout while typing English" (e.g. Dvorak) in ATOK preferences and it tracks the system input source; on Windows romaji generally follows the OS layout, but non-standard layouts still produce quirks.