Machine translation software converts text and documents between languages without a human translator, sold as a web tool for staff or as an API for developers.
This guide ranks eleven products on setup effort, the bill once documents and glossaries are counted, how your terminology leaves, and whether you can buy and cancel without a partner.
Vendors can pay for more visibility on this website, for example a higher place in a ranking or more room on a page. Commercial relationships can also affect which products we feature and how prominently. Every vendor link is marked the same, whatever the arrangement behind it. How that works.
In short
What machine translation software does
Machine translation software uses trained language models to translate text and files between languages automatically, usually offered as a web app, a plugin for translation tools, or a paid API.
Our order comes from five tests, taken in turn: effort to set up, what you really pay, what happens to your data on the way out, how locked in you are, and the intended customer. The ranking then gives places 1 to 3 to challengers outside the best-known brands, if the category has such vendors, and the incumbents follow. For why these five were chosen, and why nothing here gets a numeric score, see how we work.
Italian translation engine sold in character quotas with unlimited team users
Ranked #1 of 11 in Best Machine Translation Software in 2026.
Free tierPublished pricingEurope
Lara Translate separates what it sells by input type: text and documents are billed per character, images at 10,000 characters each, audio at 40 characters a second.
The Free plan carries 60,000 characters a month and no extra usage; Pro lifts that to 500,000. Its terms offer both a Learning Mode and an Incognito Mode, so check which one your workflow uses before loading confidential files.
What stands out
Character quota
Unlimited team users
Incognito mode
Where it costs you
Single sign-on, data residency and guaranteed service levels are Enterprise items, so mid-size regulated teams must negotiate to get them.
Document translation is metered at a minimum of 20,000 characters, so many short files cost more than their length suggests.
Right for
Teams wanting one shared character pool and monthly prices instead of per-seat billing
Wrong for
Buyers who need a self-hosted engine in their own network
ItalyFree plan with a monthly character quota; Pro and Team prices published per month; Enterprise by talking to sales
Swiss-hosted AI translator with professional linguists on call for review
Ranked #2 of 11 in Best Machine Translation Software in 2026.
Free tierPublished pricingEurope
Supertext publishes a comparison grid in which Free allows 5 file translations a month with 5 MB files, Essential 20 files at 10 MB, and Advanced 60 files at 20 MB. Single sign-on needs at least 10 users.
Enterprise adds private cloud or on-premises hosting, custom TLS certificates and an uptime guarantee. The supported list holds 28 languages, including Romansh and Swiss German.
What stands out
Hosted in Switzerland
Human review option
Translation MCP
Where it costs you
Verification by a linguist is a paid service on top of the subscription, so reviewed text costs more than the plan price shows.
File translations are capped per user each month on every published plan, and glossaries are held back until the Advanced tier.
Right for
Companies working in German, French, Italian and Swiss German who want optional human checking
Wrong for
Teams that need languages beyond the 28 on the list
SwitzerlandFree plan; Essential and Advanced prices published per month with a 30-day trial; Enterprise on request
Open source translation API you run yourself or rent by the month
Ranked #3 of 11 in Best Machine Translation Software in 2026.
Open sourceSelf-hostablePublished pricingElsewhere
LibreTranslate lists about two dozen languages on its homepage, among them Basque, Esperanto, Irish and Kabyle. Hosted tiers are Starter, Pro and Business; the entry tier bursts to 25 translations a minute and refunds within seven days.
Because the engine is Argos Translate rather than a proprietary service, you can train further language models yourself. The pages read name no company or country, so ask who stands behind support.
What stands out
Open source
Self-hosted
AGPLv3
Where it costs you
Long documents have to be split into chunks by your own code, because LibreTranslate refuses single calls above 2,000 characters.
Hosted keys are rate limited per minute, and the vendor says low-volume keys at no charge were not sustainable.
Right for
Developers who need translation inside their own network, without sending text to others
Wrong for
Buyers wanting a polished editor, glossaries and CAT tool plugins
Not stated by the vendorOpen source software to self-host at no licence cost; hosted API keys sold monthly in published tiers
Specialist translator with a free API plan and per-user team tiers
Ranked #4 of 11 in Best Machine Translation Software in 2026.
Free tierPublished pricingEurope
DeepL is a machine engine with apps around it, not a management system. That is its value: staff get document and text translation on a business contract, glossaries keep product names consistent, and the API feeds Crowdin, memoQ and Phrase.
Translation memory arrives only on the Business plan, project tracking and review comments need the Translation Flow add-on, and there is no supplier management. Budget per user for staff, and per character for the API.
What stands out
Free API plan
Per-user tiers
Translation memory on Business
Where it costs you
Translation memory only from the Business plan; workflow is a separate add-on
The free tier lacks the data terms a business needs
Right for
A company replacing staff use of free web translators with licensed accounts
Wrong for
A team that needs review workflows and supplier management
GermanyFree API plan; Individual, Team and Business published per user per month; 30-day trial on the first two; Enterprise quoted
Riga-built translation platform tuned for Baltic and other European languages
Ranked #5 of 11 in Best Machine Translation Software in 2026.
Published pricingEurope
Tilde Machine Translation prices web plans monthly or annually, with a 30 percent saving shown for annual billing, and an API plan billed per million characters plus a per-image charge. Upload limits run from 10 to 25 MB by tier.
Text translation is unlimited on every plan, so file caps are the real constraint. The API plan lists a per-image charge for image translation, which is worth confirming during the 14-day trial.
What stands out
Baltic languages
Unlimited glossaries
14-day trial
Where it costs you
The published tiers cap files per month at 5, 50 and 200, so document-heavy teams are pushed to the dearest plan.
Tilde states a focus on European languages, so test it before relying on it for other regions.
Right for
Public bodies and companies translating into Latvian, Lithuanian, Estonian and other European languages
Wrong for
Teams whose main language pairs lie outside Europe
LatviaPlans Starter, Advanced and Premium published per month; API priced per million characters; 14-day trial; Enterprise by contact
French translation engine sold per user, with server and cloud editions
Ranked #6 of 11 in Best Machine Translation Software in 2026.
Published pricingEurope
Systran has made machine translation since the 1960s and now belongs to the software group ChapsVision. Its distinguishing offer is a translation server installed in your own infrastructure, with trainable models, which suits defence, government and regulated industry.
The cloud per-user plans are published and ordinary. The on-premise route is quoted and brings hardware, upgrades and model maintenance with it, which a small IT team will feel.
What stands out
Per user
Private cloud
Model Studio
Where it costs you
Product range is wide and hard to compare
Now part of the ChapsVision group rather than independent
Right for
An organisation whose text may not leave its own network
Wrong for
A small team that just wants good cloud machine translation
FrancePro Plus and Pro Premium published per user per month with an annual-commitment discount; free trial offered
Language hub that connects translation engines to ServiceNow, Jira and Trados
Ranked #7 of 11 in Best Machine Translation Software in 2026.
Pricing on requestNorth America
Intento describes four workflows: localisation, customer experience, employee experience and a generative AI portal. Its LQA feature scores translations against your requirements and feeds automatic post-editing.
The site also hosts free courses on machine translation. The company says it handles implementation, including model evaluation and multi-agent workflows, which suggests setup is a service engagement rather than a configuration you do yourself.
What stands out
Language Hub
ServiceNow connector
Quality scoring
Where it costs you
Intento shows no prices on the pages read and leads with a demo button, so smaller teams cannot compare cost before speaking to sales.
The Language Hub is a workflow layer, so a buyer wanting only a plain text translator pays for orchestration it will not use.
Right for
Enterprises routing support tickets and localisation files through several systems at once
Wrong for
Individuals or small teams wanting a self-serve translator
United StatesSold through a demo; no prices on the pages read; the GenAI portal is billed on usage rather than seats
Language operations platform mixing AI output with an editor community
Ranked #8 of 11 in Best Machine Translation Software in 2026.
Pricing on requestNorth America
Unbabel markets real-time quality reporting, customisation and control as its core pitch, and a community of editors who review text. The site's research menu lists COMET and MAIA.
Industries named on the site include gaming, retail, tech and travel. The pages read give no price, no free plan and no trial, so a buyer should ask for those in the demo.
What stands out
LangOps platform
Human review
Helpdesk connectors
Where it costs you
Unbabel's pricing URL returned not found when fetched and its call to action is a demo, so cost cannot be checked beforehand.
Human review is part of the offer, which adds turnaround time that a pure machine API does not have.
Right for
Customer support teams translating tickets and chat inside Zendesk, Salesforce or Intercom
Wrong for
Developers wanting a pay-per-character API with public rates
United StatesSold through a demo; the pricing page returned not found; no prices on the pages read
AWS pay-per-character translation API with a twelve-month free tier
Ranked #9 of 11 in Best Machine Translation Software in 2026.
Free tierPublished pricingNorth America
Amazon Translate prices standard text, batch documents and real-time text at the same per-million-character rate, real-time Word documents at twice that rate and Active Custom Translation at four times.
Custom translation has a shorter free allowance of two months, so a pilot of Active Custom Translation must be planned inside that window. Billing by characters processed makes forecasting simple for text-only workloads.
What stands out
Pay per character
AWS service
Active Custom Translation
Where it costs you
The free tier ends twelve months after the first request, and real-time Word documents cost double the plain text rate.
Active Custom Translation costs four times the standard rate, and its free allowance lasts only two months.
Right for
Engineering teams already on AWS who need translation as one call in a pipeline
Wrong for
Office staff wanting a per-user plan with an editor
United StatesPay as you go per character; free tier for 12 months on standard and batch translation; prices published
Google API with a monthly free allowance and a document-page price
Ranked #10 of 11 in Best Machine Translation Software in 2026.
Published pricingNorth America
Google Cloud Translation lists the neural model at a per-million-character rate, a per-page document price, and custom models trained on your data. Adaptive Translation and the LLM option are priced separately from the standard model.
The free amount is a monthly credit applied to the bill, not a separate plan. The page also notes that failed batch operations are charged only for the translations that were returned, which limits the cost of retries.
What stands out
Basic and Advanced editions
Monthly free allowance
Translation LLM
Where it costs you
Batch requests are charged per target language, so translating into five languages costs five times the source characters.
Input and output characters are metered separately for the translation LLM, which makes bills harder to predict than for the standard model.
Right for
Developers who need broad language coverage and a free monthly allowance inside Google Cloud
Wrong for
Buyers who want a flat per-user subscription plan
United StatesPay as you go per character; first 500,000 characters each month free; document translation per page; prices published
Azure translation service with a two-million-character monthly free tier
Ranked #11 of 11 in Best Machine Translation Software in 2026.
Free tierPublished pricingNorth America
Microsoft Translator separates standard translation, document translation and Custom Translation, each with its own meter. The F0 tier is shared by ordinary translation and model training.
Document translation can be billed by character, and image documents by the thousand. Commitment tiers include a fixed monthly character volume with overage charged beyond it, which suits predictable workloads more than spiky ones.
What stands out
Free F0 tier
Commitment tiers
Custom Translation
Where it costs you
Custom Translation is a two-step process in which you train a model first, which costs time before any benefit appears.
Training a custom model draws on the same two million free F0 characters as ordinary translation, so a pilot can exhaust the allowance early.
Right for
Microsoft-centred organisations that want a free monthly tier plus committed pricing for volume
Wrong for
Teams wanting a seat-based plan for their staff
United StatesFree F0 tier, S1 pay as you go per character, and commitment tiers
Machine translation software uses trained language models to translate text and files between languages automatically, usually offered as a web app, a plugin for translation tools, or a paid API. Differences that change the outcome seldom sit in the features. Here is the sequence we would follow.
01
Start with pricing transparency
The split is 9 published against 2 quoted. If you want to compare numbers first, start with Lara Translate, Supertext, LibreTranslate, DeepL, Tilde Machine Translation, Systran, Amazon Translate, Google Cloud Translation, Microsoft Translator.
02
Decide how much the jurisdiction matters
Check the vendor's home market (North America 5, Europe 5, Elsewhere 1), because that is where subpoenas and data requests arrive first, whatever region holds your files.
03
Decide if open source matters
1 of these are open source: LibreTranslate. Hosting one yourself trades a subscription for maintenance.
Machine translation software: billing by character, by seat or by quota
The unit on the invoice decides what machine translation software costs. Amazon Translate, Google Cloud Translation and Microsoft Translator charge per character, so a fixed text volume is easy to forecast but a multi-language batch is not. Google multiplies characters by target languages. DeepL and Systran charge per user per month.
Lara Translate sells a monthly character pool shared by unlimited users on its Team tier. Supertext prices by subscription with file caps. Count your yearly characters, your languages and your users before comparing, because the cheapest unit rate often belongs to the plan that suits you least, and the free allowances of Microsoft Translator and Google Cloud Translation reset every month rather than expire.
Estimate yearly characters and the number of target languages
Compare per-user, per-character and pooled quota on the same volume
Check whether images, audio and documents have separate meters
Do you need an engine you can self-host or a managed service?
Self-hosting matters when text cannot leave your network. LibreTranslate is open source under AGPLv3 and runs offline, but its hosted keys cap each call at 2,000 characters, and its homepage lists about two dozen languages. Supertext lists private cloud or on-premises hosting in Enterprise, and Systran lists server, private cloud and on-premise editions.
Amazon Translate, Google Cloud Translation and Microsoft Translator are described on their pricing pages as cloud services billed by usage. Ask each vendor which tier unlocks the hosting you need before you pilot, who patches the software, what happens to support when you run it yourself, and whether the licence lets you modify it for internal use.
Write down whether text may leave your network
Ask which tier carries private cloud or on-premises hosting
Test the self-hosted engine on your own language pairs
Does the translator learn from your text, and can you switch that off?
Training on customer text is the privacy question most buyers skip. Lara Translate lets you choose between Learning Mode, which keeps texts to tune the engine, and Incognito Mode. DeepL's Individual plan lists no data training among its features. Supertext lists instant text deletion from its Essential tier upwards.
Check the contract, not the homepage, for retention periods, and ask whether an off switch is per user or per organisation. A platform that blends human editors, such as Unbabel, brings reviewers into the loop, so ask who reads your text and under what confidentiality terms, which a plain per-character API does not raise in the same way. Request the retention period in writing.
Read the terms on storing and training for the exact plan you buy
Ask whether the no-training setting is per user or per organisation
Ask who beyond the engine, such as editors, sees your text
Glossaries, translation memory and getting your terminology back out
Terminology is where machine translation software diverges. DeepL gives one glossary on Individual, five on Team, and unlimited glossaries plus translation memory only on Business. Tilde lists unlimited glossaries on every published plan, and Supertext holds glossaries back until Advanced. Systran offers Model Studio on Premium, and Intento reuses your translation history and glossaries.
Check that you can export your glossary and memory in a standard format before you load them, since moving them later is the real switching cost, and ask whether a mid-tier plan limits the number of glossaries you may keep. The ModernMT to Lara migration shows why: the vendor transfers translation memories and glossaries for you.
List the glossary and memory features needed on the cheapest qualifying tier
Test an export of your glossary and memory before loading them
Ask for the glossary limit per language pair
What goes wrong most often when buying machine translation software
Comparing a per-character API such as Amazon Translate with a per-user plan such as DeepL without converting both to your yearly volume.
Forgetting that a Google Cloud Translation batch bills source characters once for every target language you add.
Choosing an engine on the homepage demo instead of translating your own contract, manual or support ticket and having a speaker of the target language check it.
Assuming a free tier lasts, when Amazon Translate's ends after twelve months and the Lara Translate Free plan stops at 60,000 characters.
07
Frequently asked questions
8 answers
What is the best machine translation software in 2026?
Lara Translate takes first place among the 11 we reviewed. Lara Translate is the hosted service of Translated s.r.l. in Rome, and it is replacing the company's ModernMT product.
Buyers pay for a monthly pool of characters, and the Team tier lets any number of colleagues share one pool. Single sign-on, data residency and service-level guarantees sit only in Enterprise, so regulated buyers must negotiate.
Which machine translation tools publish their pricing?
9 of the 11, with the pricing model each one publishes:
Lara Translate: Free plan with a monthly character quota; Pro and Team prices published per month; Enterprise by talking to sales.
Supertext: Free plan; Essential and Advanced prices published per month with a 30-day trial; Enterprise on request.
LibreTranslate: Open source software to self-host at no licence cost; hosted API keys sold monthly in published tiers.
DeepL: Free API plan; Individual, Team and Business published per user per month; 30-day trial on the first two; Enterprise quoted.
Tilde Machine Translation: Plans Starter, Advanced and Premium published per month; API priced per million characters; 14-day trial; Enterprise by contact.
Systran: Pro Plus and Pro Premium published per user per month with an annual-commitment discount; free trial offered.
Amazon Translate: Pay as you go per character; free tier for 12 months on standard and batch translation; prices published.
Google Cloud Translation: Pay as you go per character; first 500,000 characters each month free; document translation per page; prices published.
Microsoft Translator: Free F0 tier, S1 pay as you go per character, and commitment tiers.
2 vendors do not print a price.
Is there a free machine translation tool?
Lara Translate, Supertext, DeepL, Amazon Translate, Microsoft Translator offer a free tier or a free self-hosted edition.
Where are these machine translation software vendors established?
Across 7 countries (North America 5, Europe 5, Elsewhere 1):
Lara Translate: Italy.
Supertext: Switzerland.
LibreTranslate: Not stated by the vendor.
DeepL: Germany.
Tilde Machine Translation: Latvia.
Systran: France.
Intento: United States.
Unbabel: United States.
Amazon Translate: United States.
Google Cloud Translation: United States.
Microsoft Translator: United States.
Which machine translation tools are open source?
LibreTranslate.
Which machine translation tools can you host yourself?
LibreTranslate. The remaining 10 run only on the vendor's servers.
What should you use instead of Lara Translate?
The next in line are Supertext, aimed at companies working in German, French, Italian and Swiss German who want optional human checking, and LibreTranslate, aimed at Developers who need translation inside their own network, without sending text to others.
Who should not buy Lara Translate?
Buyers who need a self-hosted engine in their own network. Single sign-on, data residency and guaranteed service levels are Enterprise items, so mid-size regulated teams must negotiate to get them..
Your machine translation software product is not among these 11? Tell us what it is for and who it serves, and we will consider it. The decision is editorial, and the ground rules are set out under software advice.
