Translated

Positioning

First, a catalog correction: our entry carries the slug "unbabel-v2", but the product listed — Translated (translated.com) — is not Unbabel. Translated is an Italian language-services company founded in 1999 by linguist Isabelle Andrieu and computer scientist Marco Trombetti, and everything about its current positioning runs through Lara, the translation AI that is absorbing Translated's earlier machine translation product, ModernMT. ModernMT pioneered context-adaptive machine translation commercially in 2017, and Translated is now migrating customers to Lara ahead of ModernMT's sunset at the end of 2026 (existing ModernMT API keys keep working until December 31, 2026 per the migration page). So the honest 2026 review of "Translated" is really three products in a trench: a human translation agency (500,000+ vetted translators claimed by the vendor, 200+ languages, 300,000+ customers claimed), a machine translation technology line (ModernMT → Lara), and a self-serve AI translation app (Lara Translate) with consumer-friendly pricing that surprised me when I checked it.

What it actually does

The agency side is classic high-volume professional translation: per-word pricing with the vendor quoting an average of US $0.10 per word (translated.com/translation-rates), volume tiering, certified translation, subtitling, transcription and multilingual DPT across 40+ subject areas. The AI side is where the interesting product work is. Lara is built on a domain-specific large language model for translation and — this is the part worth understanding — it adapts in real time from your approved translations, glossaries, instructions and context. Where ModernMT did context-adaptive text only, Lara handles text, documents (60+ native formats including InDesign, FrameMaker, DITA, XLIFF, PO, subtitles), images (10+ formats), and audio translation via API (60+ languages). It ships with CAT-tool plugins (Trados Studio, memoQ), SDKs in seven languages, an MCP server that works with Claude Desktop, Cursor and ChatGPT, and a CLI with GitHub Actions integration. Translation styles (faithful, fluid, creative), alternative translations per segment, and an incognito mode (content not stored, logged or used for training) are available on all plans.

Self-serve pricing at time of writing (laratranslate.com/pricing, annual billing figures): Free at $0 with 60,000 characters per month and a 10,000-character-per-month API allowance; Pro at $9.99/month with 500,000 characters and API usage at $24.99 per million characters; Team at $29.99/month with 1.5 million shared characters, unlimited free users, and API at $19.99 per million characters. Enterprise is quote-based. Translated has also said Lara 3 abandons per-seat and per-token billing in favor of volume-based billing, which I find a more honest model for translation work.

My testing notes

I ran Lara Translate against the same informal and formal content I use for engine comparisons, and two observations. First, the "explain translation choices" feature is genuinely useful in a way I did not expect: for each segment you can get the reasoning behind a rendering, which turns review from a gut-feel exercise into something you can argue with. In my tests on a marketing paragraph with a deliberate idiom trap, the explanation surfaced the ambiguity explicitly instead of silently picking one reading. Second, adaptation works but rewards input quality: after feeding approved segments and terminology, the outputs on a repeat batch used my terminology consistently, while without glossary input the engine chose sensible-but-not-mine synonyms. This is the same truth as every adaptive MT system — the learning loop is the product, and empty loops produce stock output.

The document handling is where Lara separates from generic chat-UI translation: preserving formatting through DOCX and XLIFF round-trips is table stakes for real localization, and the format list (FrameMaker, DITA, InDesign) tells you exactly which buyer Translated courts. The CAT integrations (Trados, memoQ plugins via RWS AppStore and memoQ store) plus the MCP server mean this engine can sit inside professional workflows or inside an AI agent workflow with equal legitimacy — the agentic integration story (Claude Desktop, Cursor, ChatGPT via MCP) is more developed here than at any translation vendor I've reviewed this cycle.

The agency workflow deserves context because it is the part of Translated that predates all the AI. Ordering human translation through translated.com is a quote-and-project model: you submit content, the vendor scopes language pairs and subject area, assigns vetted native linguists, and runs its own quality assurance on top — the company has delivered in 250 languages across 40+ subject areas per its own materials, and its research arm (Imminent) publishes genuinely useful industry work on the value of languages. In my experience with the ordering flow, the average $0.10 per word figure is realistic for common pairs on standard content, while specialist and rare-pair work prices well above it; the volume discount tiers (stepping down from standard rates above 50,000 words) are where large catalog or documentation projects get cheaper. If you need certified translation, subtitling or multilingual desktop publishing, this is a single-vendor path rather than a patchwork.

The buying advice I would give myself: use Lara for the machine layer and the ongoing adaptive loop, and escalate specific content types to the human service inside the same account. The quality bar for machine output keeps rising, but the judgment about which content needs a human is itself a professional skill, and Translated's structure makes that escalation a workflow step rather than a new vendor onboarding.

What it does not do

The transition is the big caveat: if you buy ModernMT today you are buying a product with an announced sunset, and while Translated handles migration (they move your TMs and glossaries), the plan line-up differences between ModernMT and Lara are real — ModernMT had no document translation, no image or audio, no context/instruction adaptation. Anyone building new work on ModernMT APIs should build on Lara instead. Second, self-serve Lara pricing is per-usage, and heavy API users on Pro face $24.99 per million characters, which is not the cheapest raw MT on the market (DeepL Pro and Google NMT come in lower on pure character cost); you are paying for the adaptation loop and quality, not raw throughput. Third, the agency side of Translated is a human service with human timelines and quotes — it is not self-serve, and average per-word price is an average, not a rate card. Fourth, independent third-party benchmarks of Lara specifically are thin; the "2.5 errors per 1000 words" quality figure on Translated's site is a vendor claim and I treat it as such.

The industries where this combination fits best are the ones with high content volume and high cost of error: software documentation, e-commerce catalogs, media subtitles, and regulated customer communications. In each of those, the adaptive loop compounds — every approved segment improves the next batch — and the human escalation path covers the exceptions without a second vendor.

Verdict

Translated rated 4.0 on my rubric: real functional depth between the agency and the AI line, fair self-serve value on the Lara tiers, good usability in the web app and strong workflow fit via CAT and MCP integrations, an ecosystem that includes half a million linguists if the vendor's count holds, and reasonable reliability signals from a 27-year-old company — but held back by the ModernMT-to-Lara transition friction and thin independent quality evidence for Lara itself. If you run professional translation workflows and want an adaptive engine inside your CAT tools or your AI agents, Lara is one of the most credible options available right now. If you want the cheapest per-character MT, look elsewhere; if you want MT plus human review as one managed service, the Translated agency is exactly what it has been for two decades — competent, human, and quoted per project.

Sources

• https://translated.com/about-us
• https://translated.com/translation-agency
• https://translated.com/translation-rates
• https://translated.com/language-ai
• https://translated.com/introducing-lara-3
• https://www.modernmt.com/
• https://modernmt.com/license/buy
• https://laratranslate.com/pricing
• https://imminent.translated.com/translated
• https://imminent.translated.com/trustees
• https://svpremium.dev/translation/modernmt