Connect and scale your business globally with the AI layer for language and superior quality translation across text, speech-to-speech, documents, and more 🌏💬
Thanks @jonfortt and @CNBC for having our CEO Jarek Kutylowski on to discuss our latest DeepL Voice launch. Our new models preserve more of the speaker in real time, a major step toward making voice a key interface for communicating across languages.
Introducing voice preservation in DeepL Voice. Powered by new models that carry your actual voice - not just your words - across languages in real time, including tone, rhythm, pacing and expression. Available now across 14 languages.
Hear the difference brnw.ch/21x5Hhs
We’re heading to @Salesforce#DF26. DeepL Voice will power real-time translation across 50 stages and 1,000+ sessions, and we’ll be demoing our latest Voice technology onsite. Come find us at Booth #514.
There’s an inconvenient truth that nobody in AI translation likes to talk about. What we measure in quality test scores is becoming less important than what we don’t. Gaps in MQM-type quality benchmarks may be narrowing, but the differences in real-world performance are becoming
If you’re using a general-purpose LLM for translation, quality equals latency. If you’re using a specialized model like DeepL, it doesn’t.
OpenAI GPT-5.2 is 4x slower than DeepL. Claude Opus 4.6 is 6x slower. Google Gemini 3.1 Pro is a vast 29x slower.
If you’re choosing a
