Translation CAT Tools 2026 — Trados / MemoQ / Wordfast / OmegaT / Smartcat / MateCat / Lilt / DeepL Pro / NICT VoiceTra / Rozetta T-4OO Deep Dive

Prologue — The translator's desk shook again

In summer 2025, RWS closed a deal taking a partial stake in MemoQ — the tool Kilgray built and then sold off. RWS had already absorbed SDL Trados Studio (RWS acquired SDL in 2020 for roughly $984M), Memsource (acquired 2021, rebranded as Phrase and spun into its own line), and parts of the Across customer base. Now they hold a slice of MemoQ as well.

But at almost the same time, the exact opposite trend played out. DeepL Pro shipped native plugins for Trados Studio 2024, MemoQ, Phrase, XTM, and Smartcat. GPT-4o, Claude 4.5, and Gemini 1.5 Pro became "MT engines" you can call directly from inside any major CAT through OpenAI/Anthropic/Google APIs. In other words, CAT consolidation accelerated, but the translation engines you plug into CATs went multipolar.

This article maps the 2026 CAT/translation tool landscape in a single breath. Commercial standards, open source, cloud SaaS, AI-first enterprise, LLM-based MT, file format standards, plus the local tools in Korea and Japan — and a decision matrix for who should pick what.

1. The 2026 CAT map — four categories

A CAT (Computer-Aided Translation) tool does not automate translation itself. When a translator meets a sentence they've already translated once, the CAT surfaces that earlier translation (translation memory, TM), enforces terminology (TB), and abstracts file formats (XLIFF). The MT (Machine Translation) engine is just one component inside.

The 2026 tool landscape sorts into four categories.

┌─────────────────────────────────────────────────────────────────┐
│ Commercial Desktop/Hybrid (License)                              │
│ - SDL Trados Studio (RWS)         $895~ / year                  │
│ - MemoQ Translator Pro            $770 / year                   │
│ - Wordfast Pro                    $545 / 3 years                │
│ - Across Translator Edition       Free (limited)                │
├─────────────────────────────────────────────────────────────────┤
│ Cloud SaaS (Subscription)                                        │
│ - Phrase TMS (formerly Memsource) $27~ / user / month           │
│ - XTM Cloud                       Enterprise contract            │
│ - Smartcat                        Free~$249 / month             │
│ - Crowdin / Lokalise              i18n SaaS — separate article  │
├─────────────────────────────────────────────────────────────────┤
│ Open Source (Free)                                               │
│ - OmegaT                          GPL — desktop                  │
│ - MateCat                         AGPL — web                     │
│ - Wordfast Anywhere               Free web (ad-free)             │
├─────────────────────────────────────────────────────────────────┤
│ AI-first / New (AI-first)                                        │
│ - Lilt                            Enterprise AI                  │
│ - DeepL Pro CAT integration       Plugins for every CAT          │
│ - GPT-4o, Claude, Gemini, Cohere, Reverso  LLM engines          │
└─────────────────────────────────────────────────────────────────┘

Prices are the public list as of May 2026; real enterprise contracts vary by negotiation. Over 90% of freelance translators still use Trados or MemoQ, but new translation agencies tend to look at Phrase / Smartcat / XTM first.

2. SDL Trados Studio (RWS) — the industry standard

Trados started in Stuttgart, Germany in 1984 and is the oldest commercial CAT. SDL acquired Trados in 2005, and RWS in turn acquired SDL itself in 2020 for around $984M, so the official name is now Trados Studio (RWS Trados Studio). The major 2026 release is Trados Studio 2024, with perpetual licenses retired and everything moved to annual subscription.

Core features (as of 2026):

• Cloud Capability — Trados Enterprise (the cloud TMS) integration. Local .sdlxliff files sync with cloud projects.
• Generative Translation — a unified panel for plugging in OpenAI / DeepL / Microsoft / RWS's own LLM as MT.
• AI Assistant — post-editing, terminology extraction, quality estimation (QE).
• Termbase / MultiTerm — formerly a separate license; bundled into the standard Studio subscription from 2024.
• File formats — 80+ formats: .docx, .idml, .xliff, .po, .json, .xml, .yaml, .md, .html, and more.

Pricing:

• Trados Studio 2024 Freelance Plus: $895 / year (subscription)
• Trados Studio 2024 Professional: $3,150 / year (company-wide, multi-PC)
• Trados Enterprise (TMS): negotiated separately

The perpetual license retirement was announced in September 2024 and a lot of freelancers pushed back. RWS argues that the subscription covers all the cloud features and AI, but as of May 2026, a ProZ.com survey found about 22% of Trados users say they are "considering switching tools."

Why is it still the standard:

1. TM and TB asset compatibility — 20+ years of .tmx, .sdltm, .sdltb, and .sdlxliff are scattered across every translation agency. The cost of migrating all of that exceeds the license cost.
2. LSP requirements — global translation agencies (Lionbridge, TransPerfect, Welocalize) require freelance vendors to deliver "in Trados" as the default.
3. MultiTerm ecosystem — IATE (the EU termbase), Microsoft Language Portal, and many others are distributed as .sdltb or .tbx.

3. MemoQ (RWS's competitor — and now what)

MemoQ launched in 2005 from Kilgray Translation Technologies in Budapest, Hungary. The founders had been Trados users themselves, and built MemoQ to fix what they didn't like about Trados — file-format lock-in, expensive pricing, awkward UI.

In 2025, memoQ Ltd. (Kilgray's successor) sold a partial stake to RWS. Strategic investment, not acquisition, but the industry effectively reads it as "RWS now holds a piece of its biggest competitor." That said, MemoQ remains an independent brand. RWS holds the line that "MemoQ and Trados are separate products for different workflows."

MemoQ's differentiators:

• LiveDocs — reference documents (previous translations, manuals, website dumps) that aren't aligned TMs can be searched directly as a corpus. More flexible than Trados's AutoSuggest.
• MemoQ Cloud / Server — its own TMS. Clients can work directly in a web browser via WebTrans mode.
• AGT (Adaptive Generative Translation) — introduced 2025. Automatically packages the current project's TM and TB as context and ships them to an LLM as adaptive MT.
• Quality Assurance (QA) — quantitative LQA (Linguistic Quality Assurance) scoring, plus side-by-side comparison of two translators' work.

Pricing (2026):

• MemoQ Translator Pro (freelance): $770 / year
• MemoQ Project Manager: $2,310 / year
• MemoQ Cloud Server: $2,000+ / month

Slightly cheaper than Trados with a more modern UI. New entrants tend to lean toward MemoQ.

4. Wordfast — Anywhere + Pro split

Wordfast was created in 1999 by Yves Champollion in France. It started life as a Microsoft Word macro (Wordfast Classic) and later expanded to a standalone desktop tool (Wordfast Pro) and a web tool (Wordfast Anywhere).

Current product lineup:

The fact that Wordfast Anywhere is genuinely free is still unique in 2026. No ads, the TM stays the user's property (downloadable), and MyMemory (Translated's public TM) is integrated by default. For freelancers running small projects or for students, it's the most rational starting point.

That said, big LSPs don't require Wordfast. Compatibility is guaranteed only through .tmx and .xliff.

5. OmegaT — mature open source

OmegaT is a GPL open-source desktop CAT that Keith Godfrey started in 2000. Built in Java, it runs identically on Windows, macOS, and Linux.

Where OmegaT 6.x stands in 2026:

• Free, ad-free, user-owned data — TMs and TBs are all plain text files (.tmx, .tbx) on disk.
• Git-friendly — you can commit the whole project folder to Git and collaborate that way. No other CAT is anywhere near as natural here.
• Plugins — DeepL, Google Translate, ChatGPT, and Claude are all available as community plugins.
• Downsides — the UI is 2010s-era Swing. Multi-user concurrent work is hard (Git is the workaround). Compatibility with industry termbases like MultiTerm only works through .tbx imports.

It suits students, open-source-project i18n translators, government and public sector orgs that care about data sovereignty, and freelancers who want to own their TM forever.

6. Smartcat — multi-engine plus marketplace

Smartcat spun out of ABBYY in 2016 and is headquartered in Boston, USA. As of 2026, it reports around 250,000 registered users and integrates 30+ MT engines.

Smartcat has two identities:

1. CAT/TMS — a web-based editor, TM, TB, workflow, and QA. Competes head-on with Phrase and XTM.
2. Translation marketplace — a platform to hire freelance translators directly. The buyer creates a project in the cloud, matches a translator, and processes payment all on one screen.

Pricing (2026):

• Free: $0, 1 PM seat + unlimited external translators, 2M words processed per month.
• Smartcat Pro: $249 / month (team unit, 5 user seats default).
• Smartcat Enterprise: negotiated.

The free tier is unusually generous, and Smartcat appeals to buyers who want "the whole supply chain as a SaaS." From the translator's side though, jobs received through Smartcat tend to pay less than equivalent LSP-direct work (because supply is abundant).

7. Across — the German legacy

Across Systems GmbH was founded in 1969 in Germany. The desktop (Across Translator Edition) is actually free, but the server stack (crossTank, crossTerm, crossLanguageServer) is the enterprise solution that German automotive, pharma, and legal firms use as their standard.

Why German companies love it:

• On-premises — data never leaves the corporate network. GDPR and German BDSG compliant by default.
• In-house operation — every translation step happens on your own servers.
• Workflow — optimized for large, structured documents like car manuals and pharma IFUs.

The downsides are clear. The UI is heavy, and cloud / AI integration trails the others. DeepL integration was only officially launched in 2025; Trados and MemoQ shipped the same feature in 2021.

In 2026, Across is a tool with "almost no new entrants unless an existing customer happens to bring it along."

8. XTM Cloud — the enterprise cloud done right

XTM International is a UK company, and XTM Cloud was built cloud-native from day one as an enterprise TMS. As of 2026, roughly half of the global top-100 LSPs reportedly use XTM as a backend.

Key differentiators:

• Workflow engine — visually compose complex multi-step pipelines (translate → review → DTP → QA).
• Custom connectors — direct integration with Drupal, AEM, Sitecore, GitHub, Bitbucket, Salesforce, and others.
• API-first — a fully documented REST API (swagger.json available).
• Permission model — RBAC for 100+ person organizations is well thought out.
• MT integration — DeepL, Google, Microsoft, Amazon, ModernMT, KantanMT, RWS Language Weaver — all comparable side by side.

Pricing: essentially not public. Negotiations typically land in the $50 to$150 per user per month range. Target customers are enterprises with 100+ in-house translators or large LSPs.

9. MateCat — free OSS, web

MateCat is an AGPL open-source web CAT released by Italy's Translated in 2014. Its core integrations are ModernMT (Translated's adaptive NMT) and MyMemory (a huge public TM).

The MateCat pitch:

• Completely free — no ads, every feature unlocked on signup.
• MyMemory — as of 2026 about 70 billion words of public TM are searched automatically.
• Free ModernMT tier — adaptive NMT that learns from your edits.
• Simple UI — a brand-new translator can be productive within 30 minutes.

Limitations:

• Performance is unstable on large projects. Anything over 10K segments gets sluggish.
• No custom workflows.
• Data sovereignty — everything flows through Translated's servers (anonymized contributions feed back into MyMemory; you can opt out).

A great entry point for students, translators trying things casually, or LSPs evaluating ModernMT.

10. Lilt — AI-first enterprise

Lilt is a US company founded in 2015 by two Stanford alumni. They essentially defined the "Adaptive Machine Translation" category. Where other CATs bolted MT onto an existing workflow, Lilt designed its UI from scratch around "a translator and an adaptive NMT working together."

How Lilt works:

The translator translates the first segment
   v
The model immediately learns that edit (Online Adaptation)
   v
From the next segment onward, suggestions reflect that style
   v
When the project ends, the domain adapter is saved

This loop is fundamentally different from the "TM match + post-editing" paradigm in other CATs. Lilt calls it "Real-time domain adaptation."

Product lineup (2026):

• Lilt for Translators — freelance/translator-facing. Hourly rate based.
• Lilt for Enterprise — in-house translation teams at global brands like Intel, Asics, Canva.
• Lilt Studio — Lilt's own CAT UI.

Caveat: Lilt runs its own translator pool, so it's not a "rent-the-tool-only" model for external LSPs or freelancers. Enterprise license negotiation is required.

11. DeepL Pro CAT integration + DeepL Write + DeepL Voice

DeepL is a German NMT engine published in 2017 by the Linguee team in Cologne. It started with English↔German and now supports 33 languages. In 2024 they migrated to their own LLM (DeepL Frontier), and in 2025 they officially launched DeepL Voice for live speech interpretation.

Three products in DeepL's 2026 lineup:

1. DeepL Pro Translator — general translation. Web/app/API.
2. DeepL Write — sentence-level polishing (English, German). A Grammarly competitor.
3. DeepL Voice — real-time speech interpretation (Teams/Zoom integration).

CAT integrations (as of 2026):

Pricing (2026):

• DeepL Pro Starter: 8.74 USD / month (1 seat)
• DeepL Pro Advanced: 28.74 USD / month (glossaries, CAT tool integration)
• DeepL Pro Ultimate: 57.49 USD / month (PDF/IDML translation, more glossaries)
• DeepL API: usage-based, roughly 25 USD per million characters

DeepL is strongest on European + East Asian major pairs — English to German/French/Spanish/Japanese/Korean/Chinese and the reverse. For Southeast Asian and Indic languages, Google Translate and GPT-4o tend to rate better.

12. LLM translation — GPT-4o / Claude / Gemini / Cohere / Reverso

After GPT-4 first delivered "translator-acceptable" MT quality in 2024, the 2026 LLM translation landscape looks like this.

GPT-4o (OpenAI):

• Roughly tied with or slightly ahead of DeepL on multilingual BLEU/COMET benchmarks.
• A big context window means it's strong at "translate this whole document in one go."
• API: $2.50 / 1M input,$10 / 1M output (May 2026).

Claude 4.5 (Anthropic):

• Strong long-document tone consistency. You can translate an entire book in the same voice.
• Subjectively rated slightly above GPT-4o for Korean and Japanese naturalness.
• API: $3 / 1M input,$15 / 1M output.

Gemini 1.5 Pro (Google):

• 2M token context window. You can fit an entire manual into a single prompt.
• Combined with Google Translate's vocabulary, it's particularly strong on Southeast Asian languages.
• API: $1.25 / 1M input (2M tokens),$5 / 1M output.

Cohere Command R+ (multilingual):

• Trained for balanced performance across 100 languages. Particularly strong at non-English RAG.
• One of the most price-competitive options.

Reverso:

• A French-based translation tool. Free web + Pro ($7.49 / month).
• The strength is contextual examples — search a word and you see how it's actually used in real documents.
• Popular as a study companion for students, learners, and translators.

Limits of LLM translation (2026):

1. Unstable length — translate the same sentence twice and the lengths can differ. Conflicts with CAT segment matching.
2. No factual verification — no guarantee that numbers, proper nouns, or dates carry through unchanged.
3. TM conflict — if the LLM's translation diverges from your TM's tone, post-editing cost can actually increase.

So the 2026 standard practice is "plug LLMs into the CAT as an MT engine, but always prioritize 100% TM matches first."

13. Standard formats — TMX / XLIFF / SRX / GMX / TBX

Even when you swap CAT tools, your assets must follow. Standards exist for exactly that. All are XML-based, originally maintained by LISA (Localisation Industry Standards Association) and OASIS.

TMX (Translation Memory eXchange):

• Published by LISA in 1998. Current major version 1.4b.
• The standard for moving TMs between tools.
• Every CAT imports and exports it.

<tmx version="1.4">
  <header srclang="en"
          datatype="plaintext"
          segtype="sentence"
          adminlang="en"
          o-tmf="abc"
          creationtool="MyCAT"
          creationtoolversion="1.0" />
  <body>
    <tu>
      <tuv xml:lang="en"><seg>Hello, world.</seg></tuv>
      <tuv xml:lang="ko"><seg>Hello, world.</seg></tuv>
    </tu>
  </body>
</tmx>

XLIFF (XML Localization Interchange File Format):

• OASIS standard. Current major version 2.1 (2018), with 2.2 in development.
• The standard for the translation "work in progress" — source + target + metadata (status, comments, match score) all live in one file.
• Trados's .sdlxliff, MemoQ's .mqxlz, and Phrase's .xlz are all XLIFF extensions.

<xliff version="2.1" xmlns="urn:oasis:names:tc:xliff:document:2.1"
       srcLang="en" trgLang="ko">
  <file id="f1">
    <unit id="u1">
      <segment>
        <source>Hello, world.</source>
        <target>Hello, world.</target>
      </segment>
    </unit>
  </file>
</xliff>

SRX (Segmentation Rules eXchange):

• The standard for moving sentence-splitting rules between tools.
• Rules like "the period after Mr. is not a sentence end" expressed in XML.

GMX (Global Information Management Metrics eXchange):

• The standard that makes word counts, character counts, and other quantitative metrics comparable across tools.
• Used for quoting and billing.

TBX (TermBase eXchange):

• The standard for moving termbases between tools.
• IATE (the EU terminology database) and Microsoft Language Portal both distribute as .tbx.

The 2026 reality: TMX and XLIFF 2.1 are faithfully supported by virtually every CAT. SRX, GMX, and TBX are rarely touched outside larger LSPs.

14. Korea — HanCom Translator / KIB AI Sora

In Korea, the translation tool market is effectively dominated by global CATs (Trados / MemoQ / Phrase), but two local-strength options are worth flagging.

HanCom Translator (Global HanCom):

• A translation product from Hancom (the makers of HWP).
• Translates Hancom Office (.hwp, .hwpx) files natively — no other CAT can handle .hwp directly.
• HanCom's own NMT plus a multi-engine router across DeepL, Google, and Papago.
• Main use case: .hwp document translation for the Korean public sector, defense, and financial industries.

KIB (National Information Society Agency, NIA) AI Sora Translator:

• Korean NMT released as a Korean government R&D output.
• Optimized for Korean paired with English, Japanese, Chinese, Vietnamese, Thai, and similar.
• Training data is largely Korean government public documents — natural tone for official material.
• Free web access plus a public API.

Global Korean LSPs (L&P, TransMedia, TransLine, and others) typically use Trados or MemoQ as their backend, and detour through HanCom Translator only when .hwp documents arrive — this workflow is the de facto norm.

15. Japan — NICT VoiceTra / NTT COTOHA / Rozetta T-4OO / JuliaIT

Japan has one of the richest non-English domestic NMT ecosystems.

NICT VoiceTra (National Institute of Information and Communications Technology):

• A voice interpretation app from Japan's government-affiliated NICT. Free.
• 31 languages (as of 2026) for speech-to-speech translation.
• Popularized during the 2020 Tokyo Olympics, now the standard for tourism and disaster response.
• Training data is Japanese public domain — strong on formal Japanese naturalness.

NTT COTOHA Translator:

• NTT Communications' enterprise translation API.
• Domain adaptation specialized for technical documents and financial reports.
• Frequently used as the backend for in-house translation systems at large Japanese enterprises.

Rozetta T-4OO (Translation for Onsha Only):

• Domain-specialized NMT cloud from Tokyo-based Rozetta.
• 2,000+ industry-specific models — medical, legal, patent, IT, and so on.
• Excels at "specialized Japanese domains that global translation firms can't crack."
• Japanese companies use T-4OO directly or wire it into Trados/MemoQ as an MT engine when translating to Korean or Chinese.

JuliaIT:

• An LLM-based translation SaaS from a Japanese startup.
• Uses GPT-4o/Claude as a backend with prompt tuning specialized for Japanese context.
• Launched 2025; a newer entrant but well rated on manuals and contracts.

sakuwaki:

• A community CAT for student and amateur translators in Japan. Free web CAT.
• Popular for small projects (such as manga doujin translation).

16. AI vs CAT integration — 2026 trends

The biggest shift between 2024 and 2026 isn't "will AI replace CAT" — it's "how far has AI moved inside CAT."

The early-2024 model:

[Document] -> [CAT] -> [Translator + TM match] -> [Final]
              |
              v
         [MT engine (DeepL/Google)]  <- optional

The 2026 model:

[Document] -> [CAT] -> [Translator + TM match + adaptive LLM suggestions] -> [LLM QE] -> [Final]
              |                                          |
              v                                          v
         [TM/TB assets]                  [GPT-4o / Claude / DeepL / Gemini]
                                                       |
                                                       v
                                  [Per-project domain adapter (Lilt-style)]

Five core shifts:

1. LLMs moved into the MT slot — every major CAT accepts OpenAI/Anthropic/Google/DeepL API keys.
2. Adaptive learning became the norm — Lilt's "real-time domain adaptation" spread to MemoQ (AGT), Phrase (ATA), and Smartcat (AI Translate).
3. QE (Quality Estimation) went LLM-based — an LLM pre-scores segments to flag which ones are worth post-editing at all.
4. The translator's role shifted from "generator" to "editor + QC" — the LLM produces draft zero; the human verifies and corrects.
5. Data sovereignty became a fight — pressure to disclose whether a translator's corrections feed LLM training, sharpened by the EU AI Act 2025.

17. Who should pick what — decision matrix

Freelance translator (solo):

Translation agency (LSP, 5 to 50 staff):

In-house translation team (10 to 50 staff):

Developer / i18n (software localization):

Student / learner:

References

• SDL Trados Studio (RWS): https://www.rws.com/translation/trados/
• RWS corporate: https://www.rws.com/
• MemoQ: https://www.memoq.com/
• Phrase (formerly Memsource): https://phrase.com/
• Wordfast: https://www.wordfast.com/
• Wordfast Anywhere: https://www.wordfast.com/products/wordfast_anywhere
• OmegaT: https://omegat.org/
• Smartcat: https://www.smartcat.com/
• Across: https://www.across.net/
• XTM International: https://xtm.cloud/
• MateCat: https://www.matecat.com/
• Translated (MateCat, ModernMT): https://translated.com/
• ModernMT: https://www.modernmt.com/
• MyMemory: https://mymemory.translated.net/
• Lilt: https://lilt.com/
• DeepL Pro: https://www.deepl.com/pro
• DeepL Write: https://www.deepl.com/write
• DeepL Voice: https://www.deepl.com/voice
• DeepL API docs: https://developers.deepl.com/docs
• OpenAI Platform (GPT-4o API): https://platform.openai.com/docs
• Anthropic API (Claude): https://docs.anthropic.com/
• Google Cloud (Gemini): https://cloud.google.com/vertex-ai/generative-ai
• Cohere: https://cohere.com/
• Reverso: https://www.reverso.net/
• TMX 1.4b spec: https://www.gala-global.org/lisa-oscar-standards
• XLIFF 2.1 (OASIS): https://docs.oasis-open.org/xliff/xliff-core/v2.1/xliff-core-v2.1.html
• SRX 2.0 (OASIS): https://www.gala-global.org/srx-20-april-7-2008
• TBX (ISO 30042): https://www.iso.org/standard/62510.html
• IATE (EU terminology): https://iate.europa.eu/
• ProZ.com community: https://www.proz.com/
• HanCom (Global HanCom): https://www.hancom.com/
• NICT VoiceTra: https://voicetra.nict.go.jp/
• NTT COTOHA: https://www.cotoha.com/
• Rozetta T-4OO: https://www.rozetta.jp/
• EU AI Act: https://artificialintelligenceact.eu/