“Made in Europe:” this company created the first AI architecture to rival the US and Asian models
A company that uses AI to process financial documents claims to have created the first “European” AI architecture that challenges American and Asian models.

Image by Cybernews
A company that uses AI to process financial documents claims to have created the first “European” AI architecture that challenges American and Asian models.
Lingua Custodia, a French startup, started by providing machine translation services specializing in the financial industry.
Now, branded as Dragon LLM, the company has designed the first European AI architecture, “capable of competing with American and Asian models.”
The company reveals that Dragon LLM was trained on the European supercomputers Leonardo and Jupiter. Meanwhile, language models such as ChatGPT, Claude, and Meta’s Llama are based on the Transformer architecture.
How does Dragon LLM stand out from other LLMs?
Dragon LLM was created to provide users with a sovereign, more efficient AI that consumes less energy.
The French-based LLM's release comes as Europe strives for digital sovereignty.
Cybernews has previously reported that some European countries, such as Denmark, are already shifting from US-based software by adopting Linux and open-source software.
Germany has also been on a quest to move from Microsoft by first migrating 40,000 mailboxes from Microsoft Exchange and Outlook to Open-Xchange and Thunderbird.
According to the company, the first European LLM can perform at the same level as other leading models but uses only one-third of the computational power.
The model can also run on conventional servers without needing large GPUs. The company also boasts Dragon LLM’s energy efficiency by stating that Dragon is “a sustainable alternative” that can provide faster training, requires less data, and minimizes resource use before the model even starts generating value.
Dragon LLM can also sustain double the user load compared with other models while providing users with the same quality as leading models such as Qwen or Meta.
Dragon LLM availability
The company has already released the code for its first version, Dragon-3B-Base-alpha, which was created on 3.6 billion parameters, via the Hugging Face platform for building and sharing machine learning applications.
The commercial versions should be released in the upcoming months.
With Dragon LLM, the company aims to provide users with an alternative to US-based AI models.
Unlock more exclusive Cybernews content on YouTube.