DeepL has made a significant stride in the translation world by unveiling its new and improved AI models. The goal of this innovation is to significantly enhance the translation quality for business customers. With a new generation of language models, DeepL aims to deliver translations that sound even more natural and human-like, according to a report by the “Handelsblatt.”
DeepL is a leader in artificial intelligence in Germany. According to company founder Jarek Kutylowski, the new models represent a major advancement for the company: “We are heavily investing in research to continuously push the boundaries of translation and writing quality, efficiency, and adaptability.” These new models operate similarly to the technology behind the well-known AI chatbot ChatGPT but are specifically optimized for translation tasks to avoid common AI errors.
A key advantage of the new DeepL models is their ability to provide more human-like and contextually accurate translations. This is achieved through specialized training that reduces the risk of so-called hallucinations—instances where AI generates false information. These improvements were evident in internal tests where DeepL was clearly preferred over competitors like Google, Microsoft, and OpenAI. According to Stefan Mesken, VP Research at DeepL, language experts preferred DeepL translations 1.7 times more often than those generated by ChatGPT. Additionally, the post-editing effort required to achieve DeepL’s quality was twice as high for Google translations and three times as high for ChatGPT translations.
Another key factor in DeepL’s success is the extensive proprietary data the company has accumulated over its seven-year history. This data, combined with deep expert knowledge, allows DeepL to deliver higher quality translations than many of its competitors.
With these new developments, DeepL aims to further strengthen its lead in the industry and offer its customers even more precise and natural translation solutions.
