Thread

Aleph Alpha Kolibri: How the sovereign German LLM works

413 points · 11 comments · tejaskumar__

Loading the complete thread in the background. This saved snapshot is available now. Refresh

sajithdilshan · 1791029737 · focus · HN ↗
> It knows less from memory, Multi-turn tool calling is weaker, It’s not the best coding agent
Then what does it good at? Sending faxes?
1. skrebbel · 1791032004 · focus · HN ↗
  Well that’s a pretty important skill for German users!
  1. sajithdilshan · 1791033614 · focus · HN ↗
    Unfortunately yes.
    moooo99 · 1791034631 · focus · HN ↗
    I know this is a long standing joke, but I am 27 now, born and raised in Germany, and genuinely never had the opportunity to use a Fax machine. It honestly kind of bums me out because there is something that seems cool about Fax (being fully aware that it should be fully obsolete by now)
    IshKebab · 1791035062 · focus · HN ↗
    Fax was never cool.
    hobo123 · 1791037938 · focus · HN ↗
    As a kid I interned at a bookstore, and before closing we'd write the book orders on a sheet of paper, slide it into the Fax machine to send to the book wholesale. I think the Fax printed out some kind of response a minute later.
    Next day, the books would arrive. This was maybe 1996, but I thought it was pretty cool.
    literalAardvark · 1791040841 · focus · HN ↗
    Fax was mind blowingly cool.
    Maybe you can relate to seeing webcams or Facetime when it came out or something.
    Being able to see a signed document or a picture someone took 5000km away quickly was mind boggling in its age.
    tchalla · 1791036398 · focus · HN ↗
    You have plenty of opportunities you’ve just not been forced to use them. Look into your contracts right now, I bet rhrrr are opportunities
    jfengel · 1791041148 · focus · HN ↗
    Fax was always awkward and weird. Resolution sucked and the paper handling was unreliable.
    They were never common. They were mostly business tools, and consumers never wanted them. Sometimes you'd be required to send someone a fax, and you'd have to go to an office store. (And pay rather a lot; dollars per page.)
    I hope you get to give one a try some day, for kicks, but the thrill will wear off fast.
    ketzu · 1791048498 · focus · HN ↗
    > They were never common.
    Are you sure about that? I remember we had one at home when I grew up, so I assumed they were very common at the time. (And it's not like we had this crazy business driven family use of it.)
    jfengel · 1791057964 · focus · HN ↗
    About the best data point I can cite is that I never saw one in a TV show.I don't think the Huxtables had a home fax machine. (Though maybe I just watched the wrong programs, and we're talking about a very long time ago so memory could fail.)
    ketzu · 1791105351 · focus · HN ↗
    Ah I thought it was specifically about germany.
    > I never saw one in a TV show
    Now that you mention it, I think I only remember one from the simpsons.
    But based on [1] my mom really was an outlier for getting a fax machine. Mabye we had enough contact with beurocracy to make it work it. :D Even in 1995 only 1.5 million out of 36 million (4,1%) households had one. (Although I remember it from ~2000, so too bad the data doesn't go there.)
    For the US with 14 million out of 98 million households (14%) fax machines were more popular for ordinary households than in germany. (But home-use was never the issue in the fax debate anyway, just fun facts :) )
    [1] <a href="https://www.oecd.org/content/dam/oecd/en/publications/reports/1997/01/access-to-and-use-of-information-technologies-at-home_g17a1b38/237222208123.pdf" rel="nofollow">https://www.oecd.org/content/dam/oecd/en/publications/report...
    jfengel · 1791128140 · focus · HN ↗
    It never occurred to me that anyone would try to take data on that. Way to go, OECD!
2. gizajob · 1791032180 · focus · HN ↗
  Annexing the Sudetenland.
3. stogot · 1791032227 · focus · HN ↗
  GPT was better at chat and rag before coding
4. mhitza · 1791032250 · focus · HN ↗
  I don't have the hardware to test this, but being trained to say "I don't know" instead of misleadingly talking confidently about what it does not know is interesting in itself.
  1. peri-cl · 1791033684 · focus · HN ↗
    It's not particularly good at that either, going by the "I don't know" benchmark they compared against:
    <a href="https://artificialanalysis.ai/evaluations/omniscience#omniscience-hallucination-rate-tabs" rel="nofollow">https://artificialanalysis.ai/evaluations/omniscience#omnisc... (AA-Omniscience Hallucination Rate)
    The phrasing:
    > "Of all the mixture-of-experts models Aleph Alpha compared, only Qwen3.6 35B-A3B did better, at 56.7%."
    Means they compared against only a subset of the competition. Older, tiny models with *half* their parameter count.
    Cherry-picked choice of competition, on top of a cherry-picked choice of benchmark. Not a good look.
5. gunthercuckolso · 1791033295 · focus · HN ↗
  At least its not racist unlike your orange cheetoh chud
6. m00dy · 1791033706 · focus · HN ↗
  I was thinking about the use cases for Germany residents actually, but yes, you're right. A German AI model needs to be able to handle fax machines (there's a bit variety of them) and also needs to handle letters coming home. german image-to-text.
7. pettijohn · 1791035191 · focus · HN ↗
  Seems like it's good at reading German documents and reasoning over them. Custom German-language tokenizer, and one of the least hallucinating models.
8. hasley · 1791048973 · focus · HN ↗
  I suspect there are a lot of German companies that have thousands of documents that describe their software/product. Adding or changing a feature means you have to ask 20 people whether there might something that can break some edge case behavior for which no (automatic) test exists. In this case, it would be nice to have a virtual employee who knows the content of all documents.
  I even experienced the opposite case: Some feature was first rejected since everyone agreed that it would break some behavior. After talking to several people again, I discovered that the use case which required this behavior did not exist anymore and had been officially phased out years before. Maybe, some RAG LLM could have told me, based on company documents, that there is some edge case that can be removed making the way free for the new feature.
martianvoid · 1791032212 · focus · HN ↗
I just tried to play around with it on my RTX pro 6000 setup, it spends way too many tokens on overthinking stuff even if it’s able to catch the correct approach
Its speed is pretty good on the other hand with only 3B active parameters I am getting around 170 tkn/s on fp8
1. gizajob · 1791032246 · focus · HN ↗
  > it spends way too many tokens on overthinking stuff
  Yeah. It’s a German model.
  1. rubyfan · 1791033383 · focus · HN ↗
    Hey!
  2. sajithdilshan · 1791033565 · focus · HN ↗
    This joke is either gonna get over analyzed by the Germans or gonna fly past straight over their heads
    hypfer · 1791034040 · focus · HN ↗
    I'm not sure if this is truly a good faith joke tbh.
    jamiek88 · 1791050889 · focus · HN ↗
    Jesus Christ you’re a sensitive little sunflower aren’t you? All over this thread swooning at mild jokes made mostly by Germans themselves. Chill out man. Three comments crying is enough.
    hypfer · 1791051524 · focus · HN ↗
    No, I am just standing my (and my nations) ground against what is simply bad faith disrespect and what could be described as casual racism.
    You do not get to invalidate that.
    People need to stop insulting my nation.
    dudefeliciano · 1791034547 · focus · HN ↗
    petition to make this thread the official german jokes thread for this post, it's kind of annoying how everyone feels the need to make a top level comment for their oh-so-great "germans suck" joke.
    jijijijij · 1791041712 · focus · HN ↗
    Nothing flies anywhere, we use superior high speed trains. I faxed a Humorgenehmigungsantragantrag to our Bundesunterhaltungsministerium analyst, immediately. Laugh now, but you are merely lucky the reply got delayed. As soon as the leaves on the rails are dealt with, you will get a formal response that has washed itself! Let us see who is laughing then. It is nobody!
    fph · 1791052364 · focus · HN ↗
    But, most importantly, is Bundesunterhaltungsministerium a single token in Kolibri? This could make a huge difference for the model's performance in German.
    jijijijij · 1791061115 · focus · HN ↗
    Silly you, thinking the German economy is merit-based. The only question is, if the bribe fits a single token of appreciation, to win a tailored call for bids!
  3. tfburns · 1791050170 · focus · HN ↗
    Do models from other parts of the world underthink? :P
2. tzatzikyyy · 1791034341 · focus · HN ↗
  Did you try lower reasoning modes as well?
cbarrick · 1791032737 · focus · HN ↗
I got distracted by that scroll-wheel UI component on the page. Neat!
1. reacharavindh · 1791035314 · focus · HN ↗
  The scroll wheel was cool indeed. Check their home page.. their talks on a carousel over a horizontal timeline was cool even on a phone browser.
2. dang · 1791051098 · focus · HN ↗
  Can somebody mention the URL of the page with this? We're merging the threads and I don't want a dangling pointer!
  1. cbarrick · 1791159981 · focus · HN ↗
    You have it there.
    It's on the blog post by Tejas Kumar [1], when the viewport width is narrow.
    <a href="https://tej.as/blog/aleph-alpha-kolibri" rel="nofollow">https://tej.as/blog/aleph-alpha-kolibri
    dang · 1791178232 · focus · HN ↗
    Thank you!
pythonic_hell · 1791032790 · focus · HN ↗
The benchmarks are impressive given the problem space they are working in.
1. rrr_oh_man · 1791034284 · focus · HN ↗
  German government?
  1. sajithdilshan · 1791034714 · focus · HN ↗
    More like Chancellor’s Office
driverdan · 1791033806 · focus · HN ↗
IMO the announcement is better <a href="https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/" rel="nofollow">https://aleph-alpha.com/en/blog/kolibri-has-landed-a-soverei...
1. rpdillon · 1791037307 · focus · HN ↗
  Agree, and both are on the front page. Also, Show HN is usually for your own project, but this appears to be a blog post about someone else's project (albeit "a good friend").
2. dang · 1791050847 · focus · HN ↗
  It's here: <a href="https://news.ycombinator.com/item?id=49942706">https://news.ycombinator.com/item?id=49942706. I'm currently merging the threads.
woadwarrior01 · 1791034264 · focus · HN ↗
> A bigger dense model beats it. Qwen3.8 27B ...
How is a 27B dense model bigger than a 78B MoE?
1. SyneRyder · 1791034869 · focus · HN ↗
  The 27B has all 27B as active parameters, while Kolibri is only 3.46B active parameters. I think that's what they meant anyway.
2. tfburns · 1791049031 · focus · HN ↗
  [dead]
  1. dang · 1791050945 · focus · HN ↗
    You're certainly welcome to post here! but can you please not post AI-generated or AI-edited comments to HN? It's not allowed - see <a href="https://news.ycombinator.com/newsguidelines.html#generated">https://news.ycombinator.com/newsguidelines.html#generated and <a href="https://news.ycombinator.com/item?id=47340079">https://news.ycombinator.com/item?id=47340079.
    Of course, it's impossible to know for sure what was LLM processed or not, but this post did get classified that way. That's why the software flagged it.
x1watt · 1791034369 · focus · HN ↗
Was expecting that a "sovereign" AI model would at least use their own sovereign language (German) on the website as one of the options. Anyways, all the best and happy reunification day.
1. flohofwoe · 1791034976 · focus · HN ↗
  This is a blog post, the company webpage appears to be German (at least for me: <a href="https://aleph-alpha.com/" rel="nofollow">https://aleph-alpha.com/)
  1. x1watt · 1791057019 · focus · HN ↗
    The blog is from the company itself, just noticing the irony.
    flohofwoe · 1791057264 · focus · HN ↗
    The link has changeed since I wrote that comment, the original was this:
    <a href="https://tej.as/blog/aleph-alpha-kolibri" rel="nofollow">https://tej.as/blog/aleph-alpha-kolibri
freakynit · 1791034372 · focus · HN ↗
Is it "Apple Maps bad?"
orifito · 1791035618 · focus · HN ↗
At least Germany is moving smarter than UK government...
1. testfrequency · 1791035783 · focus · HN ↗
  TIL
veryfancy · 1791035730 · focus · HN ↗
Nice to see public goods in this space.
hypfer · 1791036010 · focus · HN ↗
The ignorant, hostile, negative, and, frankly, kinda racist comments here really are just a sad showing for the currently online crowd.
But anyway. I think the main oversight when dismissing this is that not every use-case is coding a SV-style startup app. That market is quite saturated, so it would make sense to create something locally for the use-cases currently underserved by LLMs.
We will probably learn more about what this can really do once quants become available that can be run by people without an SV salary (and the biases that come with that).
CorezIoOfficial · 1791036544 · focus · HN ↗
Im surprised by how well this works. What is the difference from this and union alpha (other than the fact that it is open weights)?
pu_pe · 1791038639 · focus · HN ↗
A blog post is not a Show HN topic.
1. embedding-shape · 1791038726 · focus · HN ↗
  [delayed]
  1. pu_pe · 1791039210 · focus · HN ↗
    The author did not make this model though, they used Claude to write a blog post about it.
ThouYS · 1791038683 · focus · HN ↗
calling qwen 27B a bigger model.. I don't know man. My vram says otherwise.
rolymath · 1791039457 · focus · HN ↗
This is not a Show HN
shevy-java · 1791039572 · focus · HN ↗
Is Germany sovereign? It outsourced its defence onto the USA. Recently Trump wanted more diesel; Germany insta-submitted, also because oddly enough Macron submitted before Germany (Macron is suspicious). Before that, Leyen committed to insta-submission with a deal that made europeans poorer (and perhaps Leyen benefits from that). Canada shows the way. Many of the smaller countries in the EU too, such as Netherlands, Denmark, Finland, to some extent Sweden as well. Every time I read "sovereign" here I have to object. Nothing is sovereign here. The whole hardware is definitely not sovereign. Perhaps some of the software is, but that's about it. Plus, who gets all the data? The big US mega-corporations sniff non-stop. Remember how Facebook sniffed Libgen and Anna's Archive dry etc..., then suddenly libgen went down. The US corporations act as huge global leeches on every step of the stair. And lobbyists benefit from this too.
gilfoyle_7 · 1791040963 · focus · HN ↗
does it have GDPR compliance?
wingman-jr · 1791041282 · focus · HN ↗
While it's not perhaps clear to me that this is a true Show HN, I enjoyed the writeup and it's good to see our German colleagues across the pond taking a good shot at this. I also appreciated the brief description on the Merlin-Arthur protocol - seems like a clever way to try to tackle the "I don't know" problem.
vzaliva · 1791041286 · focus · HN ↗
"Languages: German and English" – this is odd. That means their dataset is limited. In my understanding, frontier models are trained on multilingual datasets and can combine knowledge no matter what language it was written in.
1. AnonymousPlanet · 1791056882 · focus · HN ↗
  Have you read the non-English output of those mutlilingual models? They say they put extra effort in to improve nuances and tone in this one. That alone is worth having in many settings.
  1. vzaliva · 1791072848 · focus · HN ↗
    In fact I did. I multilingual and ocassinally query ChatGPT and Claude in other languges than English.
JaggerJo · 1791042663 · focus · HN ↗
Is this a truely open source model or also open weights?
1. peterBlue75 · 1791050166 · focus · HN ↗
  The weights have been published to HF on an apache2.0 license, and the tech report is quite extensive
cheesecakegood · 1791042966 · focus · HN ↗
For those sick of “Pareto frontier” talk, just shorthand it as “it’s the best at some very particular thing”. Obviously that one thing/tradeoff it’s good at may not necessarily be compelling, but it is either a loose sign of quality, or a sign that they’ve chased some tiny edge into the ground.
I’ll be curious to see which it becomes in the next year - nba “very narrow record”, or a sign you can hang with the big boys.
Jeeetendra · 1791043069 · focus · HN ↗
3.5b active params sounds cheap until you remember all 78b still has to fit in memory. curious what the smallest practical self-hosted setup looks like for german docs.
1. layer8 · 1791043234 · focus · HN ↗
  The article talks about what setup is needed.
  1. Jeeetendra · 1791129853 · focus · HN ↗
    i meant the smallest workable setup for a few german docs, rather than the high-throughput serving setup.
tomComb · 1791043538 · focus · HN ↗
For a post to make such a big deal about sovereignty it is a bit misleading to not mention that the company is slated to be merged with Cohere, a Canadian company.
And that is a good thing - no need to hide it. Given the growing cost of keeping up, these few non-US, non-Chinese companies really need to do more sharing of efforts and costs.
Canada too is very much in need of sovereign AI options, but funding that on its own would be pretty much a waste of money. Would love to see this new German Canadian company cooperate with Mistral too, or maybe one of the Korean AI companies.
1. Loquebantur · 1791044639 · focus · HN ↗
  Given that "sovereign" these days implicitly means independence from USA and China, and Canada being spiritually in the same boat as the EU, it's pretty spot on?
  The costs of "keeping up" aren't really growing, on the contrary.
  1. user_of_the_wek · 1791052461 · focus · HN ↗
    I also read about Canada maybe joining the EU in some capacity, although I'm not sure if that was a joke. Eurovision could be a start.
    mkesper · 1791054656 · focus · HN ↗
    Ursula von der Leyen (president of european commission) propsed an "associated membership" with EU. Such an association is not defined anywhere in EU contracts yet. <a href="https://www.iss.europa.eu/publications/commentary/eus-first-associate-member-how-make-eu-canada-ties-stronger-and-tighter" rel="nofollow">https://www.iss.europa.eu/publications/commentary/eus-first-...
2. tzatzikyyy · 1791048541 · focus · HN ↗
  Hey there, I worked on Kolibri pre-training data. Agree with the point that non-US/non-Chinese labs need to pool effort which I'm also personally excited about. The merger is public, nothing to hide.
3. ivo-42 · 1791048589 · focus · HN ↗
  Hey there, I worked on Kolibri pre-training data. Agree with the point that non-US/non-Chinese labs need to pool effort which I'm also personally excited about. The merger is public, nothing to hide.
Larrikin · 1791045555 · focus · HN ↗
The name really evokes strong Kotlin library naming vibes.
erelong · 1791045725 · focus · HN ↗
is this like an unfortunate name clash with KolibriOS (kind of like how Google Gemini was a clash with the Gemini protocol project)?
1. acheron · 1791068501 · focus · HN ↗
  It’s also a name clash with arguably the finest hummingbird-based shooter for the 32X. <a href="https://en.wikipedia.org/wiki/Kolibri_(video_game)" rel="nofollow">https://en.wikipedia.org/wiki/Kolibri_(video_game)
peterBlue75 · 1791049823 · focus · HN ↗
The thing to note here, besides the transparency and the fact that it’s actually a good model that also works well on coding and agentic tasks, is that it’s the first release by a team formed less than a year ago, with a strong focus on iteration velocity. There’s more to come.
disclaimer: I‘m part of the training team, happy to answer any questions
dang · 1791050493 · focus · HN ↗
Related ongoing thread:
Kolibri: A Sovereign Open-Weight Model - <a href="https://news.ycombinator.com/item?id=49942706">https://news.ycombinator.com/item?id=49942706 - Oct 2026 (46 comments)
IndiaInfraNotes · 1791051079 · focus · HN ↗
[dead]
Ey7NFZ3P0nzAe · 1791104613 · focus · HN ↗
I'm always wondering why new models don't always adopt deepseek's KV tweaks. It's insanely valuable to have such powerful prefix caching and so cheap at inference time.
Are there drawbacks to this?

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.