German consortium releases Soofi S 30B open-source AI model

Jenia Jitsev 🏳️‍🌈 🇺🇦 🇮🇱 🇮🇷@JJitsev

JJ

@eliebakouch oh, this is what I call proper careful inspection. Very good catch, missed that. The 10x repetition given everything is rather 1-2x should be of course a trigger to look at this. Is that AIML-TUDA/QA-base indeed direct rephrasing of GPQA-D, or just similar in spirit?

10:02 AM · Jul 15, 2026

Best Crypto Cases@best_case1

BC

@effi288 @NXT4EU Herzlichen Glückwunsch zu diesem starken Launch! Es ist erfrischend zu sehen, dass aus dieser Richtung endlich etwas angeliefert wird! Ein super Start und hoffentlich bald noch weitere Erfolge!:)

6:44 AM · Jul 15, 2026

Sebastian Beyer@BeyerSebastian

Combined views

84.4K

4 posts, first seen 6h ago

783 likes55 comments

55 comments

192 saves

217 reposts

Sentiment

Positive49.1%50.9%Negative

Positive Read

Many users called Soofi S a huge win for European AI sovereignty and praised careful benchmark checks, while others dismissed it as contaminated, derivative, or a waste of time and money.

Based on 101 visible X reactions from 412 accounts.

Digg Deeper

Ask a question below.

Published answers will appear here.

Commentary on X

SB

@kimmonismus I am so excited to try it out after I sent them an email kindly requesting access and after giving a brief description of my company, explaining the industrial field of application and detailing out a possible use case idea

7:44 AM · Jul 15, 2026

Jenia Jitsev 🏳️‍🌈 🇺🇦 🇮🇱 🇮🇷@JJitsev

JJ

@eliebakouch oh, this is what I call proper careful inspection. Very good catch, missed that. The 10x repetition given everything is rather 1-2x should be of course a trigger to look at this. Is that AIML-TUDA/QA-base indeed direct rephrasing of GPQA-D, or just similar in spirit?

10:02 AM · Jul 15, 2026

Best Crypto Cases@best_case1

BC

@effi288 @NXT4EU Herzlichen Glückwunsch zu diesem starken Launch! Es ist erfrischend zu sehen, dass aus dieser Richtung endlich etwas angeliefert wird! Ein super Start und hoffentlich bald noch weitere Erfolge!:)

6:44 AM · Jul 15, 2026

ST

Stephen@0xSMW

@kimmonismus Seems like a waste of time and money. Use Gemma 4 and focus on your political policies ruining entrepreneurs.

7:21 AM · Jul 15, 2026

View all

PP

Pierre Popèrze@PPoperze

@NXT4EU Now let’s power it with cheap energy 😉

7:03 AM · Jul 15, 2026

i took a second look since this is going viral, this model is a copy paste of nemotron 3 nano, same arch, ~80% of the mixture in common which oversells very hard the sovereign aspect. but this is not even the issue, they literally train on benchmark eval set: gpqa diamond alone makes up 20% of their "capability index" and is responsible for ~70% of the gap between their model and nemotron nano (10 point difference!) so what about contamination? i looked at the mixture and they train on a dataset "AIML-TUDA/QA-base" (for 10 epochs lol) that literally is a very very light rephrasing of the gpqa diamond test set, see the screenshot 😭

9:11 AM · Jul 15, 2026

The hybrid Mamba-MoE model was trained on 27 trillion tokens.

ST

Stephen@0xSMW

@kimmonismus Seems like a waste of time and money. Use Gemma 4 and focus on your political policies ruining entrepreneurs.

7:21 AM · Jul 15, 2026

View all

PP

Pierre Popèrze@PPoperze

@NXT4EU Now let’s power it with cheap energy 😉

7:03 AM · Jul 15, 2026

i took a second look since this is going viral, this model is a copy paste of nemotron 3 nano, same arch, ~80% of the mixture in common which oversells very hard the sovereign aspect. but this is not even the issue, they literally train on benchmark eval set: gpqa diamond alone makes up 20% of their "capability index" and is responsible for ~70% of the gap between their model and nemotron nano (10 point difference!) so what about contamination? i looked at the mixture and they train on a dataset "AIML-TUDA/QA-base" (for 10 epochs lol) that literally is a very very light rephrasing of the gpqa diamond test set, see the screenshot 😭

9:11 AM · Jul 15, 2026

CH

Chubby♨️@kimmonismus

Their newly released pretraining report reveals that Soofi S 30B-A3B is based on NVIDIA’s open Nemotron 3 Nano reference architecture, the same hybrid Mamba + Transformer MoE design with ~3B active parameters and nearly identical architectural choices What’s actually new isn’t the architecture, but the training: -~27 trillion training tokens -German deliberately up-weighted -End-to-end training on Deutsche Telekom’s Industrial AI Cloud -Full training recipe, hyperparameters and evaluation methodology released openly Ngl, excited to see europe kinda trains their own models.

6:21 AM · Jul 15, 2026

"small detail" i didn't catch when first reading and change slightly the story 😂

9:13 AM · Jul 15, 2026

192 saves

217 reposts

Sentiment

Positive49.1%50.9%Negative

Positive Read

Many users called Soofi S a huge win for European AI sovereignty and praised careful benchmark checks, while others dismissed it as contaminated, derivative, or a waste of time and money.

Based on 101 visible X reactions from 412 accounts.

Digg Deeper

Ask a question below.

Published answers will appear here.

CH

Chubby♨️@kimmonismus

Their newly released pretraining report reveals that Soofi S 30B-A3B is based on NVIDIA’s open Nemotron 3 Nano reference architecture, the same hybrid Mamba + Transformer MoE design with ~3B active parameters and nearly identical architectural choices What’s actually new isn’t the architecture, but the training: -~27 trillion training tokens -German deliberately up-weighted -End-to-end training on Deutsche Telekom’s Industrial AI Cloud -Full training recipe, hyperparameters and evaluation methodology released openly Ngl, excited to see europe kinda trains their own models.

6:21 AM · Jul 15, 2026

"small detail" i didn't catch when first reading and change slightly the story 😂

9:13 AM · Jul 15, 2026