# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/kolibri-has-landed-a-sovereign-open-weight-model-49942706

# Kolibri: A Sovereign Open\-Weight Model

473 points · 286 comments

[Full discussion](<https://news.ycombinator.com/item?id=49942706>)

[Read original](<https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/>)

Category: [Models & Products](<https://hacksnap.live/?category=models-products>)

## Skept-o-meter & Hotness

Skept\-o\-meter: Low\. Estimated from 10 comments\.

2 comments for the summary\.

Peak rank: \#1

Time in Top 10: 23\.0 hours

Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\.

121 recorded rank observations from 2026\-10\-03T16:01:56\.56175\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\.

Hotness — latest 121 recorded Hacksnap ranks:

2026\-10\-03T16:01:56\.56175\+00:00: rank \#12

2026\-10\-03T17:01:36\.651036\+00:00: rank \#7

2026\-10\-03T18:01:58\.24207\+00:00: rank \#6

2026\-10\-03T19:01:19\.196589\+00:00: rank \#5

2026\-10\-03T20:00:57\.313759\+00:00: rank \#2

2026\-10\-03T21:01:04\.44439\+00:00: rank \#1

2026\-10\-03T21:20:01\.585706\+00:00: rank \#1

2026\-10\-03T22:00:15\.503126\+00:00: rank \#1

2026\-10\-03T23:00:55\.584832\+00:00: rank \#1

2026\-10\-04T08:01:00\.327459\+00:00: rank \#1

2026\-10\-04T09:00:59\.321023\+00:00: rank \#1

2026\-10\-04T10:00:42\.975336\+00:00: rank \#1

2026\-10\-04T11:01:03\.76577\+00:00: rank \#1

2026\-10\-04T12:01:45\.307985\+00:00: rank \#1

2026\-10\-04T13:00:49\.176267\+00:00: rank \#1

2026\-10\-04T14:00:56\.952543\+00:00: rank \#1

2026\-10\-04T15:00:54\.739637\+00:00: rank \#1

2026\-10\-04T16:00:51\.403054\+00:00: rank \#54

2026\-10\-04T17:00:43\.166713\+00:00: rank \#54

2026\-10\-04T18:00:46\.315535\+00:00: rank \#53

2026\-10\-04T19:00:54\.768841\+00:00: rank \#53

2026\-10\-04T20:00:51\.070442\+00:00: rank \#54

2026\-10\-04T21:00:28\.550532\+00:00: rank \#53

2026\-10\-04T22:01:36\.217651\+00:00: rank \#55

2026\-10\-04T23:01:09\.400767\+00:00: rank \#55

2026\-10\-05T08:01:51\.387003\+00:00: rank \#54

2026\-10\-05T09:01:40\.528475\+00:00: rank \#53

2026\-10\-05T10:00:56\.290772\+00:00: rank \#53

2026\-10\-05T11:01:05\.919518\+00:00: rank \#53

2026\-10\-05T12:00:56\.590881\+00:00: rank \#52

2026\-10\-05T13:01:05\.197156\+00:00: rank \#54

2026\-10\-05T14:01:04\.085718\+00:00: rank \#54

2026\-10\-05T15:00:50\.814571\+00:00: rank \#54

2026\-10\-05T16:02:09\.54139\+00:00: rank \#53

2026\-10\-05T17:00:35\.336076\+00:00: rank \#53

2026\-10\-05T18:00:20\.445491\+00:00: rank \#53

2026\-10\-05T19:02:43\.604975\+00:00: rank \#54

2026\-10\-05T20:01:09\.777459\+00:00: rank \#54

2026\-10\-05T21:01:47\.650671\+00:00: rank \#55

2026\-10\-05T22:02:59\.93099\+00:00: rank \#55

2026\-10\-06T08:02:11\.210914\+00:00: rank \#55

2026\-10\-06T08:02:14\.354697\+00:00: rank \#55

2026\-10\-06T09:01:21\.688765\+00:00: rank \#57

2026\-10\-06T10:00:58\.047773\+00:00: rank \#57

2026\-10\-06T11:00:41\.689929\+00:00: rank \#58

2026\-10\-06T12:00:41\.7705\+00:00: rank \#59

2026\-10\-06T13:01:24\.509519\+00:00: rank \#59

2026\-10\-06T14:00:26\.950817\+00:00: rank \#59

2026\-10\-06T15:00:49\.066414\+00:00: rank \#60

2026\-10\-06T16:00:58\.166747\+00:00: rank \#62

2026\-10\-06T17:01:17\.009511\+00:00: rank \#62

2026\-10\-06T18:01:43\.791429\+00:00: rank \#63

2026\-10\-06T19:00:46\.048609\+00:00: rank \#62

2026\-10\-06T20:01:14\.687294\+00:00: rank \#62

2026\-10\-06T21:02:13\.414382\+00:00: rank \#62

2026\-10\-06T22:00:40\.926271\+00:00: rank \#61

2026\-10\-06T23:03:26\.867445\+00:00: rank \#62

2026\-10\-07T08:01:41\.339804\+00:00: rank \#63

2026\-10\-07T09:02:38\.026986\+00:00: rank \#64

2026\-10\-07T10:01:35\.069531\+00:00: rank \#64

2026\-10\-07T11:02:07\.770084\+00:00: rank \#64

2026\-10\-07T12:01:14\.996697\+00:00: rank \#63

2026\-10\-07T13:00:47\.45912\+00:00: rank \#62

2026\-10\-07T14:01:18\.307739\+00:00: rank \#62

2026\-10\-07T15:01:57\.541362\+00:00: rank \#64

2026\-10\-07T16:01:28\.474022\+00:00: rank \#63

2026\-10\-07T17:01:15\.036908\+00:00: rank \#64

2026\-10\-07T18:00:53\.300079\+00:00: rank \#63

2026\-10\-07T19:01:49\.456657\+00:00: rank \#64

2026\-10\-07T20:01:59\.032487\+00:00: rank \#64

2026\-10\-07T21:05:36\.655392\+00:00: rank \#66

2026\-10\-07T22:02:27\.105241\+00:00: rank \#67

2026\-10\-07T23:01:17\.452894\+00:00: rank \#66

2026\-10\-08T08:03:55\.083808\+00:00: rank \#66

2026\-10\-08T09:03:58\.68898\+00:00: rank \#64

2026\-10\-08T10:04:05\.364235\+00:00: rank \#65

2026\-10\-08T11:02:04\.410307\+00:00: rank \#64

2026\-10\-08T12:02:36\.39618\+00:00: rank \#64

2026\-10\-08T13:03:09\.206401\+00:00: rank \#64

2026\-10\-08T14:03:30\.113384\+00:00: rank \#65

2026\-10\-08T15:02:54\.444758\+00:00: rank \#63

2026\-10\-08T16:03:13\.067807\+00:00: rank \#62

2026\-10\-08T17:03:47\.92465\+00:00: rank \#61

2026\-10\-08T18:01:50\.360431\+00:00: rank \#61

2026\-10\-08T19:02:55\.719274\+00:00: rank \#62

2026\-10\-08T20:01:36\.359022\+00:00: rank \#62

2026\-10\-08T21:03:34\.000719\+00:00: rank \#62

2026\-10\-08T22:01:13\.459732\+00:00: rank \#61

2026\-10\-08T23:01:25\.318392\+00:00: rank \#60

2026\-10\-09T08:01:53\.314021\+00:00: rank \#61

2026\-10\-09T09:02:59\.93426\+00:00: rank \#62

2026\-10\-09T10:02:22\.652527\+00:00: rank \#61

2026\-10\-09T11:02:04\.807282\+00:00: rank \#61

2026\-10\-09T12:02:16\.769295\+00:00: rank \#61

2026\-10\-09T13:02:30\.35034\+00:00: rank \#63

2026\-10\-09T14:01:34\.215134\+00:00: rank \#62

2026\-10\-09T15:01:27\.564392\+00:00: rank \#61

2026\-10\-09T16:02:06\.561056\+00:00: rank \#62

2026\-10\-09T17:03:06\.673648\+00:00: rank \#64

2026\-10\-09T18:01:02\.851052\+00:00: rank \#64

2026\-10\-09T19:01:11\.250291\+00:00: rank \#64

2026\-10\-09T20:01:32\.91907\+00:00: rank \#64

2026\-10\-09T21:02:08\.376144\+00:00: rank \#66

2026\-10\-09T22:01:19\.253075\+00:00: rank \#68

2026\-10\-09T23:01:31\.990912\+00:00: rank \#68

2026\-10\-10T08:01:00\.593376\+00:00: rank \#67

2026\-10\-10T09:01:45\.527975\+00:00: rank \#69

2026\-10\-10T10:00:53\.799353\+00:00: rank \#69

2026\-10\-10T11:00:35\.65851\+00:00: rank \#68

2026\-10\-10T12:01:11\.169647\+00:00: rank \#68

2026\-10\-10T13:00:54\.008377\+00:00: rank \#66

2026\-10\-10T14:01:43\.208766\+00:00: rank \#68

2026\-10\-10T15:01:24\.825188\+00:00: rank \#68

2026\-10\-10T16:01:10\.213328\+00:00: rank \#67

2026\-10\-10T17:00:48\.84189\+00:00: rank \#66

2026\-10\-10T18:00:57\.55132\+00:00: rank \#66

2026\-10\-10T19:00:52\.379463\+00:00: rank \#65

2026\-10\-10T20:00:21\.993912\+00:00: rank \#65

2026\-10\-10T21:01:10\.657738\+00:00: rank \#65

2026\-10\-10T22:00:50\.433569\+00:00: rank \#64

2026\-10\-10T23:01:00\.95027\+00:00: rank \#65

Aleph Alpha's Kolibri promises sovereign, IP\-safe open weights, but the supplied discussion questions whether competitive models can avoid contested training data; benchmark claims remain company\-reported\.

## The brief

Germany's Aleph Alpha has released Kolibri, an English\-German mixture\-of\-experts language model with 78B total parameters, 3B active per token, and up to 1M\-token context, published under Apache 2\.0\. The company positions it for sovereign, mission\-critical work in regulated sectors, claiming full supply\-chain control, transparency, intellectual\-property safety, and compliance with the EU AI Act and GDPR\. Aleph Alpha reports that Kolibri matches larger models on math, coding, agentic and long\-context benchmarks, and improved on internal customer\-proxy verticals, but those results are company\-reported and not independently verified\.

- Kolibri is a 78B\-parameter mixture\-of\-experts model with about 3B active parameters per token, up to 1M\-token context, and Apache 2\.0 weights on Hugging Face\.
- Aleph Alpha trained it on 768 B200 GPUs: 20T pre\-training tokens at 16k context, 3\.44T mid\-training at 64k, and 200B long\-context tokens at 256k\.
- German accounts for 21\.3% of pre\-training tokens, about 4\.3T; the company says it used translation sparingly and built a bilingual tokenizer\.
- Architecture uses 50 layers, 40 with 512\-token sliding\-window attention and 10 full\-attention layers, plus 384 experts with six active\.
- Post\-training included 268B supervised fine\-tuning tokens and reinforcement learning over more than 1\.2 million tasks, with four reasoning\-effort levels\.
- Aleph Alpha reports gains on internal customer\-proxy benchmarks across five verticals and competitive public benchmark scores, but these are company\-reported\.

## Discussion themes

Analyzed: 2026\-10\-04T09:00:30\.389842\+00:00

Analysis sample: Based on 17 of 37 usable stored comments. Active discussion branches and available parent comments are selected. The analysis input was further shortened to fit its context limit.

This sample may omit parts of the full thread. Selected themes do not measure community opinion or how common a view is.

### Open technical report and dataset documentation

Commenters praise the paper for explaining the model\-building process like a tutorial and for documenting how the dataset was made, calling the level of openness unusual\. A reply hopes this becomes a new standard and argues closed\-source approaches are hard to justify; a contributor says they worked on Kolibri pre\-training and mid\-training data and that the team strives to be open\.

Sources: [Comment 49944996](<https://news.ycombinator.com/item?id=49944996>) · [Comment 49945347](<https://news.ycombinator.com/item?id=49945347>) · [Comment 49946129](<https://news.ycombinator.com/item?id=49946129>)

### Ethical risks of democratized open\-source LLMs

A reply cautions that open\-source LLMs democratize access to an intelligence booster, which has benefits but also downsides\. It argues that ethics is not merely a lifestyle choice but a basis for social order, and that broad access means someone can misuse the tool at arbitrary scale\.

Sources: [Comment 49945632](<https://news.ycombinator.com/item?id=49945632>)

### Data cleansing at scale

In response to a contributor describing work on pre\-training and mid\-training data, a commenter asks how the data is cleansed at this scale, raising a practical data\-processing constraint\.

Sources: [Comment 49946186](<https://news.ycombinator.com/item?id=49946186>) · [Comment 49946129](<https://news.ycombinator.com/item?id=49946129>)

### Model weaknesses and practical use cases

A commenter quotes stated weaknesses such as knowing less from memory, weaker multi\-turn tool calling, and not being the best coding agent, then asks what the model is good at\. Replies extend this with a German fax joke, including one commenter saying they have never had the opportunity to use a fax machine\.

Sources: [Comment 49943617](<https://news.ycombinator.com/item?id=49943617>) · [Comment 49943840](<https://news.ycombinator.com/item?id=49943840>) · [Comment 49944031](<https://news.ycombinator.com/item?id=49944031>) · [Comment 49944144](<https://news.ycombinator.com/item?id=49944144>)

### Aleph Alpha, sovereign models, and open\-source alternatives

One commenter criticizes Aleph Alpha as having lost talent, failed to catch up, and become an investor cash grab\. A reply argues that short\-term model quality matters less than having good sovereign models over 5, 10, or 20 years; another questions why sovereign models matter if open\-source ones exist, and a further reply compares missed windows to carriage makers facing automobiles\.

Sources: [Comment 49943940](<https://news.ycombinator.com/item?id=49943940>) · [Comment 49944080](<https://news.ycombinator.com/item?id=49944080>) · [Comment 49944138](<https://news.ycombinator.com/item?id=49944138>) · [Comment 49944192](<https://news.ycombinator.com/item?id=49944192>)

### German\-language reasoning and stereotypes

Comments joke that the model 'thinks in German' means punctuality, acronyms, and committees\. Replies question whether similar nationality jokes appear elsewhere, note Deutsche Bahn as a counterexample to punctuality, and interpret the claim as the reasoning chain being based on German and its idiosyncrasies\.

Sources: [Comment 49944072](<https://news.ycombinator.com/item?id=49944072>) · [Comment 49944107](<https://news.ycombinator.com/item?id=49944107>) · [Comment 49944229](<https://news.ycombinator.com/item?id=49944229>) · [Comment 49944695](<https://news.ycombinator.com/item?id=49944695>)

## Sources & coverage

AI-generated summary · 2026\-10\-03T16:01:02\.473342\+00:00

Based on 2 of 2 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened.

Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.