DeepL introduces a language model based on language-specific LLM technology to improve corporate translation and writing proofreading performance.

DeepL, a global language AI company, announced that it has introduced a next-gen language model to its translation service. The language model is based on the Large Language Model (LLM) technology designed for translation and writing correction.

LLM-based services are available to DeepL Pro users. Starting with English, Japanese, Simplified Chinese, and German, the available languages will gradually expand. Users activate the new LLM by selecting the 'next-gen model' in the web environment.

DeepL expressed its ambition, saying, “This LLM launch is expected to upgrade the field of corporate language AI technology and establish new industry standards for translation quality and performance.”

DeepL's new solution features △Language-specific LLM, △Exclusive data for translation services, and △Model tutoring by language experts.

The language-specific LLM itself is said to provide translation and writing skills at a practical level, reducing the risk of hallucinations and mistranslations. DeepL also trains its models using proprietary data collected over 7 years solely for content creation and translation. DeepL employs thousands of language experts who directly tutor language models to improve and maintain quality.

DeepL revealed that in its own blind test conducted with language experts, the experts preferred the DeepL solution's translation results 1.3 times more than Google Translate, 1.7 times more than ChatGPT-4, and 2.3 times more than Microsoft.

“This announcement is just the beginning of DeepL’s journey to deliver enterprise-ready LLM-based language AI solutions,” said Jarek Kutylowski, Founder and CEO of DeepL. “As a research-driven, growing company, DeepL is investing heavily in pushing the boundaries of quality, efficiency, and reach in translation and writing. Our ultimate goal is to enable our global customers to more easily scale and succeed across the globe, without language barriers.”

“While the launch of our new LLM is a significant milestone for us, our ultimate goal is to deliver demonstrable impact to our clients,” said Stefan Mesken, VP Research at DeepL. “When professional translators tested our new model alongside our existing LLM and competitor models, they found that it took twice as much editing with Google Translate and three times as much with ChatGPT to get the same quality translation as with our new model.” He also emphasized, “This means that DeepL’s solution can have a significant impact on the efficiency and productivity of companies that need to translate large volumes of documents, allowing employees to spend time on more important tasks.”

DeepL announced that the results of a blind test conducted by leading linguists showed that the translation performance of the new LLM was about 1.7 times better than the existing model for English, Japanese, and Simplified Chinese combinations in its translation service. In addition, the performance was improved by 1.4 times for English and German combinations.

Meanwhile, DeepL launched DeepL for Enterprise, a customized language technology product, in June.