Top
Best
New

Posted by NotInOurNames 10/28/2025

EuroLLM: LLM made in Europe built to support all 24 official EU languages(eurollm.io)
773 points | 606 commentspage 4
trilogic 10/28/2025|
Great job, Thank you.

We support your work and offer backup and distribution. Here a copy just in case: https://hugston.com/uploads/llm_models/EuroLLM-22B-Instruct-...

aurintex 10/28/2025||
Is it planned to have a VLM or something compareable like Qwen3-VL for the future?
jug 10/28/2025|
A multimodal release is planned.
fulafel 10/28/2025||
See also Apertus: https://www.swiss-ai.org/apertus
Steen3S 10/28/2025||
If multi‑lang is the goal, why not translate the output of the big labs?
sublimefire 10/28/2025||
Surely that would need to be both input and output. But even then you could easily get lost in translation as the intent in one language might mean slightly different thing in another. Thus you could get subpar results.
layer8 10/28/2025|||
Because there is always something lost in translation.
ideasarecool 10/29/2025||
Languages don't operate one to one.
rmoriz 10/28/2025||
Maybe we can call it "open weights" and not open source?
elias_t 10/28/2025||
Are there any benchmarks that exist for those 24 languages?
ks2048 10/28/2025||
The detailed results are in appendix to the paper: https://arxiv.org/abs/2506.04079
nodja 10/28/2025|||
It's on the huggingface readme

https://huggingface.co/utter-project/EuroLLM-9B#results

https://huggingface.co/utter-project/EuroLLM-9B#english

moralestapia 10/28/2025||
dupe of https://news.ycombinator.com/item?id=45733832

which sank to the bottom thanks to HN's invisible hand

Oh wait, one's not supposed to notice

morkalork 10/28/2025||
It's more like the default is to be ranked near the bottom unless your comment gets traction during the brief window of time it is ranked first for being new. Seeing your comments go splat after that window expires is not some nefarious conspiracy..
moralestapia 10/28/2025||
Oh, you'd be surprised to know what's behind many of those "conspiracies"!
Zufriedenheit 10/28/2025||
EU officials should create an environment where abundant private companies can afford to put out many great open models instead of funding some selected individuals with taxpayer money.
bogtog 10/28/2025||
They report benchmarks on the huggingface page (https://huggingface.co/utter-project/EuroLLM-9B)

They almost exclusively compare their model to prior models from 2024 or older and brag about "results comparable to Gemma-2-9B". I'm not sure what I expected. The eurollm.io homepage states "EuroLLM outperforms similar-sized models", which just seems like a lie for all practical purposes

An overly charitable interpretation is that EuroLLM isn't a reasoning model and has minimal post-training, so they sought out comparisons to such models (they're still ignoring reasoning models that have non-reasoning modes)

aeontech 10/28/2025|
> They almost exclusively compare their model to prior models from 2024

As another comment here noted, the title is missing (2024) - this model was released almost a year ago, last December, so it's not surprising that that's the models they compare to.

websku 10/28/2025|
I'm looking to try this for ActorDO
More comments...