October 28, 2025
Europe speaks AI—comments scream louder
EuroLLM: LLM made in Europe built to support all official 24 EU languages
EU AI speaks 24 languages; comments ignite benchmark brawls, grammar raids, and US vs EU debates
TLDR: EuroLLM is a European, open-source AI that speaks all 24 EU languages, with a strong 9B model and a smaller 1.7B for devices. The community cheered the 9B benchmarks, roasted the 1.7B, nitpicked the title, and argued US vs EU dominance—making AI sovereignty the hot topic.
Europe just dropped EuroLLM, a homegrown AI that speaks all 24 official EU languages, trained on a mega dataset and out now as open source on Hugging Face. Voice and vision are promised next, and there’s a lighter 1.7B version for phones and gadgets. But the real party? The comments. One user turned the thread into a linguistics pub quiz, listing every EU language and noting that Maltese is the only Afro‑Asiatic one, which spawned trivia jokes and flag emojis. The loudest chant was “Benchmarks?”—and when numbers surfaced, the crowd split: the 9B model got serious props, while the 1.7B was dunked on as “meh.” Cue debates over edge devices vs big-boy models. A grammar cop stormed in to correct the title (“all 24 official EU languages”), triggering a mini‑meme war of pedants vs vibes. Meanwhile, a realist shrugged, “It’s just another Horizon grant,” cooling the hype with funding cynicism. Then came the geopolitical heat: why do the biggest models still come from the US and China, with Europe lagging (Mistral aside)? That fired up AI sovereignty talk, with fans cheering this as a rare EU flex—powered by the MareNostrum 5 supercomputer and EuroHPC high‑performance computing—plus a cameo for researcher André Martins, the Professor Power‑Up Europe needed.
Key Points
- •EuroLLM introduces an open-source multilingual LLM suite supporting all 24 official EU languages, with future vision and voice capabilities planned.
- •The flagship EuroLLM-9B model has 9B parameters, trained on 4+ trillion tokens across 35 languages; Base and Instruct variants are released.
- •A 1.7B-parameter model, trained on similar data, is designed for edge-device deployment.
- •The project was developed by a European consortium and trained on the MareNostrum 5 supercomputer, with support from Horizon Europe, the ERC, and EuroHPC.
- •EuroLLM is fully open source, available on Hugging Face, and claims to outperform similar-sized models.