Best Machine Translation Software

DeepL is a Germany based neural machine translation service offering a free web translator alongside DeepL Pro plans for individuals, teams, businesses and enterprises. It translates text and full documents including Word, PDF, PowerPoint and Excel files while preserving formatting, and supports customization through glossaries, style rule lists and, on higher tiers, translation memory. It is available as a web app, desktop apps for Windows and macOS, mobile apps for iOS and Android, browser extensions, and a developer API.

Current official pricing (billed annually) is Free at $0/month for 50,000 characters, Individual at $8.74/month, Team at $28.74 per user/month, and Business at $57.49 per user/month, each raising character limits, file translation counts and adding admin, SSO or translation memory features; Enterprise is custom priced. A Write Pro writing assistant add-on costs $7.49 per user/month. Paid plans include a 30-day free trial and guarantee no data is used for training.

Amazon Translate is a neural machine translation service from AWS that developers call via a simple REST API to translate text and documents in real time or in batch. It supports 75 languages and over 5,550 translation combinations, with automatic source language detection and support for TXT, HTML, DOCX, XLSX, PPTX, and XLIFF files while preserving formatting.

Pricing is pay-as-you-go per million characters, with new AWS accounts getting 2 million characters free per month for the first 12 months. Customization options include Active Custom Translation using your own parallel data and Custom Terminology for brand and industry-specific terms, plus controls for formality, brevity, profanity masking, and do-not-translate tags.

Systran is a machine translation company offering Translate Pro, a subscription service supporting 55+ languages and 140 language combinations, with text, file (Word, Excel, PowerPoint, PDF, JSON, XML), and webpage translation. The cloud service is hosted in Europe and complies with GDPR, with translated data neither stored nor reused. A free basic web translator is also available.

Paid plans are billed per user per month, with Plus and Premium tiers differing mainly in monthly page allowance and added features like CAT tool and TMS connectors, plus access to SYSTRAN Model Studio for custom models. Enterprise deployments with dedicated infrastructure and custom pricing are available for larger organizations through Systran's parent company, ChapsVision.

Unbabel is a Language Operations (LangOps) platform that combines neural machine translation with human editors to help enterprises deliver multilingual customer support and content at scale. Its AI quality estimation gives teams visibility into translation performance, and workflows can be tuned for speed, quality, or cost depending on market or content type. The platform integrates with tools such as Zendesk, Salesforce, Intercom, and Freshdesk, and its proprietary TowerLLM model powers translation across dozens of languages.

Founded in Lisbon, Portugal in 2013, Unbabel was acquired by TransPerfect in August 2025 and now operates as a division within TransPerfect. It continues to serve enterprise customers such as Adidas, Patagonia, and Skyscanner, with pricing available only through custom quotes based on volume and requirements.

ModernMT is an adaptive neural machine translation engine built by Translated that learns in real time from human corrections and uses document-level context, drawing on up to 100,000 words of surrounding text, to improve translation accuracy. It integrates with CAT tools such as Matecat, memoQ, and Trados Studio, offers an open source core on GitHub, and provides an API for custom workflows across roughly 200 languages.

ModernMT is being retired: the company states it is evolving into a new product called Lara, with ModernMT set to sunset by the end of 2026. New account registrations on modernmt.com are closed, existing API keys keep working until December 31, 2026, and customers are being migrated to Lara plans. Historical pricing was volume based, ranging from about $8 to $50 per million characters depending on the service tier.

Azure Translator translates text and documents instantly or in batches across more than 100 languages, delivered through a single API that also handles source language detection. Detection matters more than it appears in real deployments: user-generated content, support tickets and inbound messages rarely arrive labelled, so a translation pipeline that requires the source language to be specified fails at the first step.

The most interesting capability is the hybrid approach to models. Alongside neural machine translation, Microsoft offers flexible translation using foundation models such as GPT-4o for tone, gender and style customisation. Gender is a genuinely difficult problem in machine translation, since many languages mark gender where English does not, and a system guessing wrongly produces output that is fluent and incorrect in a way that is embarrassing rather than merely inaccurate. Being able to control it explicitly is a real capability rather than a refinement.

Customisation covers domain-specific terminology, letting an organisation define how terms are translated based on previously translated documents and build customised translation models without machine learning expertise. That last clause is the commercially relevant part: model customisation is usually gated behind data science capability most organisations do not have.

The agentic use cases named are specific and current: translation for call centres, multilingual conversational agents and in-app communication. Real-time translation inside a support conversation is a different engineering problem from translating documents, so naming those separately reflects genuine capability rather than positioning.

It is delivered within Microsoft Foundry Tools as a production-ready, scalable engine. Pricing follows Azure consumption based on characters translated.

Google Cloud Translation covers websites, apps, documents and media across 189 languages, and the language count is the first thing to weigh: for organisations serving genuinely global audiences, coverage of the long tail of languages is usually the constraint rather than quality in the major ones, where most providers are now adequate.

Three approaches are offered rather than one, and choosing between them is the real decision. Pre-trained models work instantly with no setup and suit general content. Custom translation models are trained on your own material, which matters where terminology is specialised: a legal, medical or engineering translation that uses the wrong term for a concept is worse than useless regardless of how fluent it reads. Gemini-powered Adaptive Translation is the newest route, aimed at nuanced accuracy.

Adaptive translation is the genuinely different capability here. Rather than applying a fixed model, it adapts output using examples of your preferred style and terminology, which addresses the persistent problem that machine translation is fluent but tonally wrong for a given brand or document type. Support for custom large language models extends this further.

Everything is reachable through the Cloud Translation API, so it integrates into applications and content pipelines rather than being a standalone tool. That is the appropriate shape for translation, which is almost always a step inside another process rather than a destination.

Documentation, release notes, a translation course and Architecture Center guidance are published openly. Translation can be tried free, and pricing follows Google Cloud consumption based on characters processed, with new customers receiving platform credits.