Machine translation stopped being a product you sell around 2024. It is now a line item inside tools agencies already pay for, and the question shifted: why pay per word when a model can do it in seconds?
The work moved instead. Deciding what a machine should never touch, keeping a brand voice intact across languages and proving the output is publishable are now the expensive parts — the parts vendors charge for.
Here is what that tooling costs in 2026, priced from the vendors’ own pages: just the numbers an agency needs to compare.
What changed in the localisation stack
A translation agency’s stack has three parts: the TMS where work is queued, machine translation engines, and the human review layer. AI has moved into all three — routing jobs to the right engine, grounding output in your own translation memory and glossaries, and scoring quality before a linguist opens a file.
That is why prices vary so widely. Some vendors bundle AI into a per-seat subscription, some sell it as a monthly add-on, and some price everything per word — turning the AI layer into a margin question, not a software question.
1. Lokalise — the AI-native platform for product localisation
Lokalise was built for software strings — app interfaces, product copy, websites — rather than legal or marketing documents, which is where AI routing pays off fastest.
Three tiers are published: Explorer at $149 a month, Growth at $379 and Advanced at $1,049, with a monthly/yearly toggle on the page. Enterprise is sales-only. AI is part of the subscription, not a separate line: Lokalise lists context-aware scoring, multi-LLM smart routing and custom AI profiles built on the agency’s own assets.
Why it made the list: the clearest example of AI bundled into a per-seat subscription, so AI work needs no second invoice.
2. Crowdin — the cheapest credible entry point, with AI on a prepaid balance
Crowdin is where small agencies and open-source-adjacent teams tend to start, and the free tier is usable for public projects rather than a two-week trial.
Published pricing is Free for public projects, Pro from $75 a month and Team from $450 a month, Team on annual billing only at $5,400 a year. Crowdin AI and its AI Pipeline handle auto-translation, proofreading draws on a prepaid Managed Balance, and engines such as DeepL sit inside the plan as add-ons.
Why it made the list: the lowest entry price here, with a prepaid balance that keeps AI usage predictable.
3. Phrase — the broadest platform, and the steepest price jump
Phrase covers both halves of the market — Phrase Strings for software content and Phrase TMS for documents — under one login.
The Freelancer plan is $27 a month billed annually for one TMS seat, Software UI/UX is $525 a month billed annually, and Team is $1,245 a month billed annually with unlimited TMS seats and 20 Strings seats. Phrase Studio, its video subtitling and dubbing layer, is a volume-based add-on, so media work is priced separately from documents.
Why it made the list: the freelancer tier is the cheapest route onto an enterprise-grade CAT tool.
4. Transifex — the plan with a quantifiable AI word allowance
Transifex sits between the developer-first tools and the enterprise suites, and is common in product localisation where a small team needs the workflow without a long implementation.
The plan table sits behind a contact button, but the page’s FAQ publishes the numbers: Starter from $135 a month billed annually, or $160 billed monthly; Growth from $200 a month billed annually, or $236 monthly, including 60,000 AI words a year on annual plans.
Why it made the list: the 60,000-word AI allowance is the most measurable AI bundle here.
5. Smartling — the per-word ladder that shows where the margin sits
Smartling sells translated output as much as software: the platform routes work to vendors and tracks quality, so its price list is per word rather than per seat.
Its plans page sets out the whole ladder: machine translation starting at $0.0075 per word, AI translation at $0.06, AI human translation at $0.12 and human translation at $0.20, with AI Hub for high-volume automated translation and the platform subscription quote-only.
Why it made the list: the full machine-to-human ladder in one place is the clearest view of what AI changes about unit economics.
6. Wordbee — the only vendor that prices the agentic layer on its own
Wordbee is a long-standing TMS used by mid-size language service providers and in-house teams, with project management depth agencies running many vendors need.
Infinity Enterprise is €150 per power user a month billed annually, and Agentic AI is an annual add-on that layers onto any Translator plan: Agentic Pro at €450 a month, Agentic Business at €1,350 and Agentic Enterprise at €3,150, all billed annually with unlimited users, or a 90-day pilot at €299 a month; the add-on runs on your own LLM keys.
Why it made the list: the only vendor with a published, standalone price for the agentic layer — a reality check on what AI costs at the top of this market.
7. memoQ — the traditional CAT licence, with AI as an unpriced add-on
memoQ is translator-first rather than client-first: translation memory, term bases, quality assurance and a desktop editor linguists tend to like.
Starter is €2,750 a year, or $3,025 in USD, billed annually, including one project manager licence and five linguist subscriptions; every tier above it is quote-only. Its AI feature, memoQ Adaptive Generative Translation, is described as something you can add to any plan — with no published price.
Why it made the list: a flat annual licence, no usage metering, no published AI number — the honest baseline for AI-native platforms.
8. DeepL — the machine layer most agencies already resell
DeepL is not a TMS and does not pretend to be: it is the engine inside many agency workflows, and usually the cheapest line in the stack.
Individual is €7.49 a month billed annually, Team €24.99 per user a month and Business €49.99 per user a month, all including VAT; Enterprise is custom. Write Pro, its AI writing assistant, is an add-on at €6.99 per user a month.
Why it made the list: at €25–€50 a seat it is the cheapest part of a modern stack — and what clients compare your per-word rate against.
Also worth knowing
Weglot publishes the most transparent website-localisation pricing here: Starter €15, Business €29, Pro €79, Advanced €299 and Extended €699 a month, with AI translation included. Tolgee prices at €58, €133 and €373 a month with 400,000 MT credits included.
Bureau Works now serves at wxrks.com and publishes a seven-tier ladder from Free and Starter at $38 a month up to Pulsar at $3,300. RWS sells Trados from £24 a month for the browser and desktop editions, with Language Weaver Pro as a paid add-on. For media localisation, ElevenLabs runs from $6 to $990 a month, priced in credits.
And two free options worth knowing: Matecat is an open-source CAT tool with adaptive machine translation built in, powered by ModernMT — which no longer lists a price.
What is quote-only, and why it matters
Three vendors publish nothing at all. XTM markets an AI localisation platform with an agentic layer but asks you to request a personalised quote, Plunet shows “request a quote” on every tier, and XTRF is now folded into XTM.
usually the figure depends on volume or seats
How to choose without sitting through five demos
Convert each to a per-word cost for your volume
Second, match the pricing shape to the business: per-seat for a stable in-house team, per-word for variable client volume, a monthly AI add-on for an agency selling a premium AI-assisted service that must appear on the invoice.
Third, test on your own assets. These platforms differ most in how well they use the translation memory and glossary you already own, so bring 2,000 words of approved work and one messy client file.
The honest takeaway
AI has not made translation cheap — it made the first draft nearly free and the review layer more important. The most useful number here is Smartling’s ladder: the gap between $0.06 and $0.20 per word still needs a person.
Treat an AI price without a number as a question for the demo
Cover photograph: FrDr, CC BY-SA 4.0, via Wikimedia Commons.