DeepL secured a $2 billion valuation following a $300 million investment in May 2024. This growth stems from a foundation laid in 2008 when Gereon Frahling and Leonard Fink established Linguee in Cologne. Jaroslaw Kutylowski led the machine translation team at Linguee starting in 2016 and assumed the CEO role in July 2019. The company used a 5.1-petaflop supercomputer at Verne Global in Iceland to train its neural networks using hydroelectric and geothermal energy. This infrastructure allowed DeepL to launch its translator on August 28, 2017.
The team built the system using a massive corpus of bilingual sentences scraped from the web, including European Union and public authority texts. This data allowed the proprietary neural machine translation models to outperform competitors in blind tests at launch. I find the technical shift from statistical models to neural networks very clear in their early success. While Google Translate launched in 2006, DeepL arrived with a system that specifically targeted linguistic nuance through human-guided training. The company’s success in Japan was rapid, as it became the 30th most popular website in the country within 17 months of release. In March 2021, the company added 13 European languages, such as Bulgarian, Czech, Danish, Estonian, Finnish, Greek, Hungarian, Latvian, Lithuanian, Romanian, Slovak, Slovenian, and Swedish. They later added Indonesian and Turkish in May 2022, Ukrainian in September 2022, and Korean and Norwegian in January 2023.
Expanding the AI toolset
The company shifted its business model from B2C to B2B with the March 2018 release of DeepL Pro. This product provides API access and software plug-ins for tools like SDL Trados Studio. DeepL Pro also includes enterprise-grade security like TLS encryption and text deletion to ensure that submitted text is not saved on the server.
| Product | Primary Use | Primary Function |
|---|---|---|
| DeepL Translator | Machine Translation | Human-guided accuracy |
| DeepL Pro | Enterprise Translation | No text storage on servers |
| DeepL Write | Text Improvement | Grammar and style suggestions |
| DeepL Voice | Speech Translation | Real-time meeting captions |
| DeepL Agent | Workflow Automation | Autonomous application operation |
DeepL expanded its writing assistant, DeepL Write, by adding Spanish and French in the summer of 2024. Earlier, in January 2024, the tool gained support for Italian and Portuguese. In January 2026, the company added Luxembourgish and Irish. DeepL Voice launched in November 2024 and expanded to include voice-to-voice translation in more than 40 languages in April 2026. The company also released DeepL Agent in November 2025 to operate business applications autonomously.
I note that the company scale grew from 43 employees in 2019 to over 1,000 by May 2026. The number of full-time staff doubled between 2019 and 2020, growing from 43 to 86 employees. This workforce includes engineers and over 1,000 contract-based human translators who assess quality. The company serves over 200,000 business customers, including many Fortune 500 companies. DeepL is currently considering a US IPO with a market capitalization of up to $5 billion.
Challenging the giants
DeepL targets the enterprise market by providing translations for companies like Roche, Fujitsu, and Siemens. I find their strength in the ability to handle specific terminology through custom glossaries. This feature helps marketing teams localize ads and retailers track international regulations, while it also allows companies to define custom terms to avoid errors. However, the product is not always perfect. One user reported that the system once translated the German word "Post" as "the German post office" in a context that actually referred to the Austrian post office.
The company’s market position remains strong despite having fewer languages than its rivals. While Google Translate supports 108 languages, DeepL launched with only 26. DeepL claims its users are 30% more likely to prefer its model over Google Translate and 130% more likely to prefer it over Microsoft’s software. I observe that many professionals prefer DeepL for its ability to capture nuance and tone in complex texts. A 2025 research paper introducing the WMT24++ benchmark found that frontier large language models, such as OpenAI’s o1, Anthropic’s Claude, and Google’s Gemini, ranked above traditional machine-translation systems like DeepL across the languages tested.
Can the company maintain this lead as large language models from Google and Microsoft become more capable? You should consider if the specialized nature of DeepL’s models provides enough of a gap to withstand the scaling of general-purpose AI. DeepL remains a massive presence in the European tech scene, especially in Cologne.
