🌏 The machine translation newsletter - September 2026

Cohere launched North Small Translate, an open machine translation model with 218B parameters (25B active) and a 16k context length that the company says outperforms other open MT models as well as Google Translate and DeepL across 50+ languages. DeepL Voice added real-time speaker identity and tone preservation for live meetings across 12 languages, and Google launched Gemini 3.8 Live with real-time speech handling for speakers switching languages mid-conversation. GPT-6 Astra leads machine translation benchmarks but remains below human-level translation quality, according to tests by RWS and the Last Translation Benchmark.

Hi folks Translation AI is coming to Tech Week Speakers from Google, Microsoft, DeepL, Meta, ModelFront, Krisp and more will meet up in person in San Francisco at this unique event bringing together key translation AI providers and enterprise localization leaders. Live translation is getting more personal this month. DeepL Voice now preserves who’s speaking and how they sound, and Google ’s Gemini 3.8 Live handles speakers switching languages mid-conversation. On the model side, GPT-6 Astra is leading translation benchmarks, and Cohere ’s new open model, North Small Translate , claims to beat both DeepL and Google Translate . Enjoy — and share 🗞️ News - GPT-6 Astra https://www.reddit.com/r/machinetranslation/comments/1wcfcu0/gpt6 astra tested in translations/ leads machine translation benchmarks https://www.reddit.com/r/machinetranslation/comments/1wcfcu0/gpt6 astra tested in translations/ , but remains below human-level translation quality, according to tests by RWS and the Last Translation Benchmark . - The newly implemented EU AI Act https://www.reddit.com/r/machinetranslation/comments/1wk1q4k/eu ai act exempts translation workflows from ai/ does not require translation tools to disclose AI use, unlike text generators and rewriters. - HarperCollins ’s new manga line, KAZE https://www.animenewsnetwork.com/news/2026-08-23/kaze-us-manga-imprint-plans-to-launch-with-5-8-titles-in-2027-to-grow-to-25-30-titles-per-year-by-/.240866 , will not use AI translation, in support of human translators. - Anthropic adds SynthID-Text https://www.anthropic.com/news/claude-text-watermark watermarking https://www.anthropic.com/news/claude-text-watermark to Claude ’s text output, which is kept when the text is copied or reformatted, but not when it’s translated or rephrased. - Google limits its AI sign language https://slator.com/google-ai-sign-language-tool-low-stakes-communication/ translation tool https://deepmind.google/blog/putting-sign-language-ai-into-users-hands/ to casual conversation, citing accuracy and liability concerns. 🚀 Launches - Cohere launches North Small Translate https://huggingface.co/CohereLabs/North-Small-Translate-1.0 , an open machine translation model with 218B parameters 25B active and a 16k context length that outperforms other open MT models as well as Google Translate and DeepL across 50+ languages . - DeepL Voice launches real-time https://www.reddit.com/r/machinetranslation/comments/1wjzfa3/deepl voice launches realtime speaker identity/ speaker identity https://www.reddit.com/r/machinetranslation/comments/1wjzfa3/deepl voice launches realtime speaker identity/ and https://www.reddit.com/r/machinetranslation/comments/1wjzfa3/deepl voice launches realtime speaker identity/ tone preservation https://www.reddit.com/r/machinetranslation/comments/1wjzfa3/deepl voice launches realtime speaker identity/ for live meetings, keeping individual vocal characteristics and emotional delivery across 12 languages . - Google Translate launches iOS https://www.reddit.com/r/machinetranslation/comments/1wjzjqv/google translate rolls out ios and android/ and https://www.reddit.com/r/machinetranslation/comments/1wjzjqv/google translate rolls out ios and android/ Android https://www.reddit.com/r/machinetranslation/comments/1wjzjqv/google translate rolls out ios and android/ upgrades https://www.reddit.com/r/machinetranslation/comments/1wjzjqv/google translate rolls out ios and android/ : Android users can keep translating while using other apps or with the screen locked, and iPhone users can hold the phone to their ear to listen to the translation privately. - Google launches Gemini 3.8 Live https://www.reddit.com/r/machinetranslation/comments/1wk0zov/google launches gemini 38 live with realtime/ , with real-time speech-to-speech reasoning and automatic language detection across 97 languages for code-switching during live dialogue. 🛠️ Community launches - LST — Local Subtitle Translate https://www.reddit.com/r/machinetranslation/comments/1whf4is/quick update lst is now live on the chrome web/ launches on the Chrome Web Store , using Ollama to translate Netflix subtitles locally in real time. - i18n-keyless https://www.reddit.com/r/machinetranslation/comments/1wh7yft/ive built an i18n library and automatic/ , an open-source internationalization library with automatic translations, skips the usual setup of translation keys and per-language files. - Manga Translator https://www.reddit.com/r/machinetranslation/comments/1wdbq5x/i made a tool that translates manga pages while/ detects speech bubbles, runs OCR, translates into 30+ languages , and re-renders text back into the original bubbles. - AITnovel https://www.reddit.com/r/machinetranslation/comments/1w68gyn/i made an ai translation service specifically for/ , a paid AI translation service for entire novels, supports TXT , EPUB , and PDF , with editable glossaries for names and terminology. - ALST https://www.reddit.com/r/machinetranslation/comments/1w1jgd5/alst opensource realtime android screen/ , an open-source Android app, translates on-screen text and overlays the translation directly on top of the original in real time, using Gemini Vision with an offline backup mode. - Champollion https://www.reddit.com/r/machinetranslation/comments/1w0yuhn/open machine translation development suite/ , an open evaluation and development suite for low-resource languages , aims to support community projects like Masakhane and AmericasNLP . - cinemacc-subtitle-skill https://www.reddit.com/r/machinetranslation/comments/1vuk206/ai kept making the same mistakes when translating/ , an open-source add-on for AI assistants like Claude , builds a glossary of names and terms before translating subtitles , to keep them consistent across a file. - An open-source https://www.reddit.com/r/machinetranslation/comments/1vtkf9e/i built an opensource video translation dubbing/ video dubbing https://www.reddit.com/r/machinetranslation/comments/1vtkf9e/i built an opensource video translation dubbing/ pipeline https://www.reddit.com/r/machinetranslation/comments/1vtkf9e/i built an opensource video translation dubbing/ combines Whisper transcription, translation, and VoxCPM2 voice cloning, demoed on an English-to-Turkish dub. - ConveyThis open-sources a GitHub https://www.reddit.com/r/machinetranslation/comments/1wjdizi/open source github for multilingual websites/ repo for translating multilingual websites https://www.reddit.com/r/machinetranslation/comments/1wjdizi/open source github for multilingual websites/ and a Claude skill https://www.reddit.com/r/machinetranslation/comments/1vz4pm0/i built a claude skill for translating entire/ that translates static sites while preserving HTML structure, SEO metadata, hreflang, canonicals, and sitemaps. - An AI news app https://www.reddit.com/r/machinetranslation/comments/1wc88a3/i built an ai news app that translates live news/ translates live news from 150+ countries , summarizes 2,000-word articles into 3 bullets , and compares coverage across 60,000 publishers . 📚 Research - A comparison between a general-purpose model, https://www.reddit.com/r/machinetranslation/comments/1wiv6m1/comparing three translation models on a budget gpu/ Qwen3-8B https://www.reddit.com/r/machinetranslation/comments/1wiv6m1/comparing three translation models on a budget gpu/ , and two dedicated translation models, https://www.reddit.com/r/machinetranslation/comments/1wiv6m1/comparing three translation models on a budget gpu/ HY-MT1.5-1.8B https://www.reddit.com/r/machinetranslation/comments/1wiv6m1/comparing three translation models on a budget gpu/ and https://www.reddit.com/r/machinetranslation/comments/1wiv6m1/comparing three translation models on a budget gpu/ TranslateGemma-4B https://www.reddit.com/r/machinetranslation/comments/1wiv6m1/comparing three translation models on a budget gpu/ , on a budget GPU https://www.reddit.com/r/machinetranslation/comments/1wiv6m1/comparing three translation models on a budget gpu/ finds the dedicated models faster, but less reliable at preserving numbers and code. - The Last Translation Benchmark https://www.reddit.com/r/machinetranslation/comments/1w6zxr8/machine translation is not solved and it may take/ paper collects 3,456 hard-to-translate examples that break state-of-the-art models, arguing machine translation is far from solved. - A year translating novels with local AI models https://www.reddit.com/r/machinetranslation/comments/1w54jbe/translating whole novels with local llms what a/ yields lessons on keeping glossaries, splitting the work across two models, and why invented words still break machine translation. - A developer mixing two languages in the same sentence finds that hinting https://www.reddit.com/r/machinetranslation/comments/1wdbbv0/why does whisper drop half of any sentence when/ Whisper https://www.reddit.com/r/machinetranslation/comments/1wdbbv0/why does whisper drop half of any sentence when/ ’s API with three candidate languages https://www.reddit.com/r/machinetranslation/comments/1wdbbv0/why does whisper drop half of any sentence when/ , not the two actually spoken, stops it from dropping words. - AMTA publishes a framework for evaluating translation https://www.reddit.com/r/machinetranslation/comments/1vugl6o/amta publishes framework for evaluating/ QE https://www.reddit.com/r/machinetranslation/comments/1vugl6o/amta publishes framework for evaluating/ systems https://www.reddit.com/r/machinetranslation/comments/1vugl6o/amta publishes framework for evaluating/ . - RWS TrainAI launches M-GATE https://www.rws.com/artificial-intelligence/train-ai-data-services/research/m-gate/by-model/ , a benchmark of 70 AI models across 30 languages , finding top reasoning models often score worse than a random guess on non-English grammar. - DATAmundi tests AIDA https://slator.com/datamundi-reports-translation-benchmark-results/ multi-agent frameworks on terminology-constrained translation, evaluating strict glossary adherence for enterprise MT. ❔ Questions - Why is Reddit https://www.reddit.com/r/machinetranslation/comments/1whijwh/why is reddits translation feature such a mess/ ’s translation feature https://www.reddit.com/r/machinetranslation/comments/1whijwh/why is reddits translation feature such a mess/ such a mess? Can’t they plug in other translation APIs? - Are Meta ’s OMT-NLLB https://www.reddit.com/r/machinetranslation/comments/1wgra2a/are omtnllb metas omnilingual mt arxiv 260316309/ Omnilingual MT model weights publicly released anywhere? - How should I position myself https://www.reddit.com/r/machinetranslation/comments/1w8zf3h/nontech background 20 years crosslingual work and/ for a job search non-tech background , 20+ years of cross-lingual work, two years diagnosing LLM translation failures ? - Which small local LLMs https://www.reddit.com/r/machinetranslation/comments/1w8lqvt/best small llms for translations/ work best for translation? - What’s the best AI tool for translating dense https://www.reddit.com/r/machinetranslation/comments/1vugsiy/best ai tool to translate dense/ philosophy https://www.reddit.com/r/machinetranslation/comments/1vugsiy/best ai tool to translate dense/ and https://www.reddit.com/r/machinetranslation/comments/1vugsiy/best ai tool to translate dense/ psychoanalysis https://www.reddit.com/r/machinetranslation/comments/1vugsiy/best ai tool to translate dense/ books https://www.reddit.com/r/machinetranslation/comments/1vugsiy/best ai tool to translate dense/ ? - How do I translate a DRM-protected https://www.reddit.com/r/machinetranslation/comments/1vtdjee/aiuto per tradurre un file epub/ EPUB https://www.reddit.com/r/machinetranslation/comments/1vtdjee/aiuto per tradurre un file epub/ from English to Italian ? - Is Google Translate https://www.reddit.com/r/machinetranslation/comments/1vsz3tw/how was google translates voice recorded/ ’s voice https://www.reddit.com/r/machinetranslation/comments/1vsz3tw/how was google translates voice recorded/ recorded word by word, or broken down into phonetic sounds? - Is live translation https://www.reddit.com/r/machinetranslation/comments/1vs1wh7/looking for feedback is live translation during/ during meetings https://www.reddit.com/r/machinetranslation/comments/1vs1wh7/looking for feedback is live translation during/ actually useful? OLVA team asking for feedback - Is there a multilingual https://www.reddit.com/r/machinetranslation/comments/1vrgxfu/i wish there was a multilingual voice/ voice transcriber https://www.reddit.com/r/machinetranslation/comments/1vrgxfu/i wish there was a multilingual voice/ for WhatsApp ? - How do you handle documents that need https://www.reddit.com/r/machinetranslation/comments/1vz3sn6/how can i handle documents that need to be/ re-translating https://www.reddit.com/r/machinetranslation/comments/1vz3sn6/how can i handle documents that need to be/ every time they’re updated? - DeepL ’s quality has dropped. Is Google Translate https://www.reddit.com/r/machinetranslation/comments/1vy1yjs/deepl quality drop now google translate is better/ better now https://www.reddit.com/r/machinetranslation/comments/1vy1yjs/deepl quality drop now google translate is better/ ? - Which rule-based https://www.reddit.com/r/machinetranslation/comments/1vus01t/rulebased machine translation/ machine translation https://www.reddit.com/r/machinetranslation/comments/1vus01t/rulebased machine translation/ system supports Portuguese to English ? 📆 Events - Translation AI https://www.reddit.com/r/machinetranslation/comments/1wjzikq/translation ai event at tech week/ at https://www.reddit.com/r/machinetranslation/comments/1wjzikq/translation ai event at tech week/ Tech Week https://www.reddit.com/r/machinetranslation/comments/1wjzikq/translation ai event at tech week/ — 8 October — San Francisco 🇺🇸 - WMT 2026 https://machinetranslate.org/wmt26 — 28 - 29 October — Budapest 🇭🇺 💻 Jobs - Senior Software Engineer https://www.reddit.com/r/machinetranslation/comments/1vx8k4b/senior software engineer localization on ai/ , Localization on AI Answers https://www.reddit.com/r/machinetranslation/comments/1vx8k4b/senior software engineer localization on ai/ at Google — Belo Horizonte 🇧🇷 - AI Ops, Localization https://www.reddit.com/r/machinetranslation/comments/1wcg29e/ai ops localization at workday pleasanton ca/ at Workday — Pleasanton 🇺🇸 - Localization Product Strategist https://www.reddit.com/r/machinetranslation/comments/1wmohy3/localization product strategist at google boulder/ at Google — Boulder 🇺🇸 - Senior Product Manager https://www.reddit.com/r/machinetranslation/comments/1wmokda/senior product manager language translation/ , Language & Translation Platform https://www.reddit.com/r/machinetranslation/comments/1wmokda/senior product manager language translation/ at Medallia — McLean 🇺🇸 - Applied Scientist II https://www.reddit.com/r/machinetranslation/comments/1wmolci/applied scientist ii genaillm translation/ , GenAI/LLM, Translation Services https://www.reddit.com/r/machinetranslation/comments/1wmolci/applied scientist ii genaillm translation/ at Amazon Science — Seattle 🇺🇸 - Applied Scientist/Research Engineer https://www.reddit.com/r/machinetranslation/comments/1wmomds/applied scientistresearch engineer llm training/ , LLM Training Data https://www.reddit.com/r/machinetranslation/comments/1wmomds/applied scientistresearch engineer llm training/ at Propio Language Services — Overland Park 🇺🇸