Languages supported by popular machine translation engines in 2026

Language coverage in machine translation has changed substantially since 2023. The practical question used to be simple: does this engine support my language pair? Today it has a second layer: are you looking at a dedicated NMT engine, which publishes verified per-pair counts and maintains consistent quality across those pairs, or an LLM-based tool, which supports broad multilingual capability but does not publish language counts in the same way and varies more in quality by pair?

This article covers the current language support for the main NMT engines and LLM-based translation tools, what those counts mean in practice, and how to read them when making a tool selection.

In this article

1. How to read language coverage numbers
2. Google Translate: 249 languages
3. Microsoft Translator: 130+ languages
4. Amazon Translate: 75 languages
5. DeepL: 33 languages
6. ModernMT: 200 languages
7. LibreTranslate: open-source, self-hosted
8. LLM-based tools: a different kind of coverage
9. Language coverage comparison table
10. How to choose based on language needs
11. FAQs

How to read language coverage numbers

A language count on its own tells you less than it looks like. Two things are worth understanding before using these numbers for a tool decision.

First, there is a difference between the number of languages an engine officially supports and the quality of translation across those languages. Google Translate supports 249 languages, but quality varies considerably between a high-resource pair like English to French (abundant training data, consistently strong output) and a low-resource pair like English to Igbo (far less training data, more variable output). A high number does not mean uniform quality across all of them.

Second, NMT engine language counts are generally API-verified: the provider has tested that translation requests for those language codes return usable output. LLM language support works differently. Models like ChatGPT, Claude, and Gemini can translate dozens to hundreds of languages, but they do not publish verified per-pair counts in the same way, and quality variation across languages is higher and less predictable.

With that framing, here is where the main tools stand.

Google Translate: 249 languages

Google Translate is the broadest-coverage machine translation service publicly available. As of March 2026, it supports 249 languages and language varieties, following its 2024 expansion of 110 new languages. That expansion added coverage for more than 614 million additional speakers, with roughly a quarter of new languages coming from African language families. In late 2025, Google upgraded Translate with its Gemini model, improving output quality on conversational content, idioms, and slang across major pairs.

For most standard professional language pairs, Google is a reliable default choice. Its weaknesses appear in less common language pairs and in content types that require strong cultural adaptation or domain-specific terminology.

Supported languages: 249 (including language varieties)

Strongest pairs: Major European pairs, Chinese, Japanese, Korean, Arabic, and most high-resource languages

Where to check: translate.google.com lists supported languages; Google Cloud Translation documentation lists API-supported language codes

Microsoft Translator: 130+ languages

Microsoft Translator supports more than 130 languages and dialects and is part of Azure Cognitive Services. It integrates natively with Microsoft Office, Teams, and SharePoint. In January 2025, Microsoft launched Translator Pro for enterprise, adding customised phrasebooks, expanded language coverage, and availability in US Government cloud. At $10 per million characters, it is the lowest-cost major NMT API, and its free tier (2 million characters per month with no expiration) is the most generous among the main providers.

It performs well on standard business content across its supported languages, with particular strength in Asian language pairs and real-time translation use cases where low latency matters.

Supported languages: 130+

Strongest pairs: Asian languages (Chinese, Japanese, Korean), European pairs, real-time scenarios

Free tier: 2 million characters per month, no expiration

Amazon Translate: 75 languages

Amazon Translate supports 75 languages and dialects. It is designed primarily for AWS-native workflows: it integrates directly with S3, Lambda, and other AWS services, and is optimised for high-volume, automated pipelines. Its Active Custom Translation feature allows organisations to supply domain-specific parallel data to improve output consistency for their content.

Coverage breadth is lower than Google and Microsoft, but for teams already operating in the AWS ecosystem and working in major language pairs, it is a practical and cost-efficient option.

Supported languages: 75

Strongest pairs: Major global language pairs; strong performance on Asian languages in independent benchmarks

Free tier: 2 million characters per month for the first 12 months

DeepL: 33 languages

DeepL supports 33 languages, including Hebrew and Vietnamese added in 2024. It has the narrowest language coverage among the main commercial NMT providers, but consistently produces the most natural-sounding output for European language pairs. In MachineTranslation.com's internal benchmark across 5,000 words of mixed technical and marketing content, DeepL scored 94.2% accuracy (the highest among standalone engines tested) with output described as the most human-sounding for French and Spanish specifically.

DeepL launched DeepL next-gen in 2024, a purpose-built LLM specifically for translation. It supports the same 33 languages as the classic model and improves quality on longer texts. For teams primarily working in European languages and prioritising naturalness over coverage breadth, DeepL remains the quality benchmark for that category.

Supported languages: 33

Strongest pairs: European pairs, particularly German, French, Spanish, Italian, Dutch, Portuguese

What it lacks: Languages outside its core European and East Asian set; no coverage for most African, South Asian, or Middle Eastern languages

ModernMT: 200 languages

ModernMT is an adaptive machine translation engine that learns from corrections over time, which makes it progressively more accurate for domain-specific content the more it is used within an organisation. It supports 200 languages and offers document-level translation, which means it considers the full document context rather than translating segment by segment.

It is less widely known than the four engines above but is a practical option for teams with large volumes of repetitive-structure content (product catalogues, technical documentation, customer support) that benefit from adaptive learning.

Supported languages: 200

Best use cases: High-volume, repetitive content; teams that want a system that improves with use; document-level context handling

LibreTranslate: open-source, self-hosted

LibreTranslate is an open-source translation engine that can be self-hosted on your own infrastructure. It currently supports around 50 languages. Output quality is generally below the commercial NMT engines for most language pairs, but for organisations with strict data residency requirements or privacy constraints who cannot send content to external APIs, self-hosted LibreTranslate is a usable option.

Its language support is limited and quality is inconsistent, particularly on less common pairs. For professional translation workflows, it is best understood as a privacy-first fallback rather than a primary quality option.

Supported languages: ~50

Best use cases: Data-sensitive workflows, self-hosted infrastructure, research environments

LLM-based tools: a different kind of coverage

LLMs like ChatGPT (GPT-4.1), Claude (Opus 4 / Sonnet 4), and Gemini (2.5 Pro / Flash) translate text as part of their general-purpose capability. They do not publish verified per-language pair counts in the same way NMT engines do, and their coverage works differently.

In practice, LLMs handle broad multilingual translation reasonably well across dozens of major languages. For high-resource language pairs where there is abundant training data, they produce output that often matches or exceeds dedicated NMT engines on quality benchmarks for nuanced or domain-specific content. For lower-resource languages, quality is more variable and less predictable than with NMT engines that have specifically trained on those pairs.

The practical implication: if your language pair is a major global language (European, Chinese, Japanese, Korean, Arabic), LLMs are a viable option and often produce stronger output on complex content. If your language pair is low-resource, a dedicated NMT engine with verified support for that pair is the safer starting point.

LLMs also introduce the standard single-model limitation that applies to NMT engines and is discussed below: without a cross-check mechanism, there is no signal when the output contains an error.

Language coverage comparison table

How to choose based on language needs

If your language pair is major and well-supported across all providers (English to French, Spanish, German, Chinese, Japanese, Korean), the coverage decision is easy — all main engines support it, and the real question is quality and cost.

If your language pair falls outside the core European or East Asian set, the order to check is: Google first (249 languages, highest coverage), then Microsoft (130+, particularly strong in Asian languages), then Amazon (75, AWS-native). DeepL is the last resort for non-European language pairs given its 33-language limit.

For low-resource languages specifically, Meta's NLLB model (open-source, trained on 200+ languages) is worth considering. In MachineTranslation.com's internal testing, Meta's NLLB engine achieved 15% higher accuracy than Gemini on low-resource languages including Kinyarwanda and Lao.

For professional, client-facing, or regulated content in any language pair, the more useful question is not which engine supports the language but whether the output can be trusted without verification. All single-engine tools share the same limitation: there is no internal signal when the translation contains an error. As MachineTranslation.com's internal analysis shows, the errors that remain in 2026 are almost entirely semantic rather than syntactic (wrong tone, wrong register, wrong term) and no single engine can reliably catch its own mistakes of that kind.

Running translations through multiple engines and comparing outputs is one way to address this. MachineTranslation.com aggregates 22 models including all the NMT engines in this article, runs them simultaneously, and uses its SMART system to surface the consensus output alongside quality scores for each. For users working across multiple language pairs who want to test coverage and quality in one place, it is a practical way to do that without managing separate integrations. Free to use, no sign-up required.

FAQs

1. Which machine translation engine supports the most languages?

Google Translate, with 249 languages as of March 2026. It expanded by 110 new languages in 2024, adding coverage for more than 614 million additional speakers, with particular growth in African language families.

2. How many languages does DeepL support?

33 languages as of 2026. DeepL added Hebrew and Vietnamese in 2024. Its coverage is narrow compared to Google, Microsoft, and Amazon, but it consistently produces the highest-quality output for European language pairs among dedicated NMT engines.

3. Is there a machine translation engine that supports all languages?

No engine supports all languages. Google is the closest at 249. For languages outside what major commercial engines support, Meta's NLLB open-source model covers 200+ languages and performs particularly well on low-resource language pairs. LibreTranslate covers around 50 languages and can be self-hosted.

4. What happened to IBM Watson Language Translator?

IBM discontinued the Watson Language Translator service in June 2024. Teams using it need to migrate to an alternative. Google Cloud Translation, Microsoft Translator, or Amazon Translate are the most direct replacements depending on language pair and infrastructure requirements.

5. Do LLMs support more languages than dedicated MT engines?

LLMs like ChatGPT, Claude, and Gemini have broad multilingual capability that in practice extends to many more languages than dedicated NMT engines list. However, they do not publish verified per-pair language counts, and quality across lower-resource languages is less consistent than dedicated NMT engines that have trained specifically on those pairs. For major global languages, LLMs often produce stronger output on complex content. For uncommon language pairs, a dedicated NMT engine with explicit support is usually the safer starting point.

6. How can I test which engine performs best for my language pair?

MachineTranslation.com runs 22 models simultaneously including Google, DeepL, Microsoft, Amazon, ChatGPT, Claude, Gemini, and others, showing the output and quality score for each. Running your actual content through all of them in one place is a faster way to evaluate coverage and quality for your specific language pair than testing each API separately. Free, no sign-up required.

Compare language coverage and quality across 22 AI models at MachineTranslation.com — free, no sign-up required.