🌏 The machine translation newsletter - May 2026

May 15, 2026

Hi folks! The machine translation newsletter is back! The amazing progress in machine translation has only accelerated.

Modern machine translation (learned from data, not based on hand-coded rules) started exactly 20 years ago, with the original launch of Google Translate.

Now trillions of words are translated every month. (Not hyperbole, that’s the actual number.)

The quality available has improved fantastically, by any measure. (Model evaluation is as unsolved as ever.) Even free, open-source models now translate at quality better than the best APIs a decade ago, between thousands of languages.

Convenient built-in translation features are now standard across top platforms like Reddit, X and Whatsapp. Book fans are hacking together their own tools, now that machine translation can do context across sentences and take prompts.

Progress is even coming to the high-value content, where manually checking the AI is the bottleneck. Enterprise localization teams are using AI to automagically confirm millions of segments, while keeping human quality.

To help us all keep track, we’ve launched new sections at machinetranslate.org, like models and automatic post-editing.

🚀 Launches

Language support

Google Translate now supports more than 247 languages.
DeepL now supports 109 languages, after adding more than 70 in November.
Meta released Omnilingual MT for 1,600 languages in a free, open-source model.
Machine Translate now lists 615 languages.

Voice and video

Apple integrated system-wide Live Translation into FaceTime, phone calls and messages via Apple Intelligence.
Apple launched real-time in-person translation directly through AirPods Pro.
Google Translate launched real-time speech translation, then speech-to-speech with Gemini.
DeepL launched voice translation in November 2024, then real-time voice at its Spring 2026 event.
Microsoft launched a Live Interpreter API, and Microsoft Teams added turn-based translation for multilingual meetings.
Translated launched Lara for iOS.
Ray-Ban Meta and Huawei shipped translation in smart glasses and on-device.
Zoom launched Voice Translator for real-time multilingual meetings.
YouTube rolled out AI auto-dubbing with “Expressive Speech”.
Instagram launched AI dubbing for Reels that clones voices and syncs lip movements.

Consumer apps

WhatsApp added native, on-device message translation in 19+ languages.
X / Grok now translates and recommends foreign-language posts.
Amazon launched Kindle Translate, letting self-published authors AI-translate books into multiple languages.
SafePal launched in-app machine translation.

MT ⇄ LLMs

Google Translate integrated Gemini capabilities.
Microsoft brought LLM-powered translation to Azure Translator with a September 2025 preview launch.
Translated added reasoning model Lara Think and 200-language support to Lara.
DeepL launched DeepL Agent and Clarify, and updated its API with a new language model and AI writing tools.
OpenAI’s ChatGPT launched a translation app page.
OpenAI launched GPT‑Realtime‑Translate, a live translation model that translates speech while keeping pace with the speaker.

Open models

Google released TranslateGemma, an open suite of translation models.
Alibaba released Qwen-MT, a dedicated machine translation model supporting 92 languages.
Cohere released Command A Translate and the open-weight Tiny Aya for 70+ languages.
Tencent open-sourced a 440MB translation model that runs fully offline on smartphones, supporting 33 languages.
Machine Translate now covers models and lists 11 providers.

Quality estimation and automatic post-editing

ModelFront announced general availability of automatic post-editing.
Advanced localization teams, like Booking.com, have shifted to using AI to verify AI scaling translation while keeping human-level quality.
Amazon’s LLM translation playground uses RAG to inject translation memory into prompts, improving BLEU scores without fine-tuning.
Machine Translate now supports automatic post-editing providers and lists 9 QE integrations and 4 APE integrations across TMS platforms.

More product and feature launches →

📚 Research

Google Translate released the WMT24++ and SMOL datasets for high- and low-resource languages.
FUSE, a new MT evaluation metric for indigenous languages.
Pearmut, a tool for human evaluation of translation.
WMT 2025 results were released at EMNLP.

More research news →

📈 Business

Reddit credited machine translation with driving record international user growth.
DeepL cut 250 jobs amid AI-driven restructuring.
ElevenLabs grew to $500M ARR and raised more investment.
Unbabel exited to Transperfect.
RWS outsourced AI to Cohere.