HTTP 200 verified twice
Finding your next stop…
Loading the latest directory information.
Loading the latest directory information.
Aleph Alpha is a Heidelberg-based European AI company that designs, develops, and operates sovereign, domain-specialized large language models (SLLMs) on EU infrastructure for enterprises, public institutions, and defense, using its Savanna model-factory pipeline and explainability methods such as AtMan.
Start here for the decision-making essentials: what Aleph Alpha does, who it is for, how it is accessed, and the first-party sources behind this profile.
See official pricing
Read-only public captures of Aleph Alpha’s homepage. Screenshots are dated, never live embeds, and open full-screen.
Short answers to the questions buyers and builders commonly ask about Aleph Alpha. Each answer cites the shared ledger below, where every source is listed once.
Domain-specialized language models that connect domain knowledge with internal workflows i · Digital watermarks and clear provenance for AI-generated content · Real-time output evaluation using external Judge-LLMs (Prüf-Modelle) · AtMan (attention manipulation) explainability method published at NeurIPS 2023
Die Technologie von Aleph Alpha schließt diese Lücke und verbindet auf einzigartige Weise die Wissensarbeiter mit ihren operativen Aufgaben in sensiblen Umgebungen.
Enterprises, public institutions, and defense organizations in industry, public sector, an · Critical enterprises and public institutions
Unsere Kunden sind Spezialisten in Industrie, öffentlichem Sektor und Verteidigung, wo Daten und Expertise souverän bleiben und Ergebnisse zuverlässig sein müssen
Designing, developing and operating customized, sovereign large language models (SLLMs) fo
In enger Partnerschaft mit unseren Kunden konzipieren, entwickeln und betreiben wir individualisierte, sichere Sprachmodelle (SLLMs) für reale, geschäftskritische Umgebungen und Entscheidungen.
Models run exclusively on European infrastructure and are trained per applicable EU law
Unsere SLLMs laufen kompromisslos auf europäischer Infrastruktur und werden gemäß dem geltenden EU-Recht speziell auf Ihrer Domäne trainiert.
Open-source system for German federal AI infrastructure permitting
Konzipiert als Open-Source-System mit dem Ziel, die Genehmigungszeiten für große Infrastrukturprojekte zu reduzieren
Multi-phase training pipeline including pre-training, supervised fine-tuning (SFT), and re
Modernes Post-Training besteht zum Beispiel aus einer Supervised-Fine-Tuning-(SFT)-Phase gefolgt von einer Reinforcement-Learning-(RL)-Phase.
This profile connects the jobs Aleph Alpha is described as handling with its delivery model, access options and the subjects used to match it to related products in this directory.
Each fact points to a recorded source, making it easy to distinguish verified product information from claims that need checking.
HTTP 200 verified twice
Aleph Alpha
Aleph Alpha develops specialized language models and AI solutions for enterprises and publ · Headquartered in Heidelberg with offices in Berlin, Bayreuth, and Munich · Operates across four German locations in Europe
Aleph Alpha develops specialized language models and AI solutions for enterprises and public institutions in Europe
Heidelberg, Germany (Speyerer Straße 14, 69115 Heidelberg), with additional offices in Berlin (Wallstraße 16, 10179 Berlin), Bayreuth (Weiherstraße 26, 95448 Bayreuth), and Munich (Einsteinstraße 174, 81677 München)
Europe, with operations across four German locations
Enterprises, public institutions, and defense organizations in industry, public sector, and science where data sovereignty and regulatory compliance are required
Designing, developing and operating customized, sovereign large language models (SLLMs) for business-critical domains including law, administration, industry, and science
Models run exclusively on European infrastructure and are trained per applicable EU law
European Union (models trained in accordance with applicable EU law)
AI assistants for public administration; AI agents for extracting verifiable insights from complex/sensitive documents in semiconductor and automotive industries; Requirements Engineering for RFQ processing
Domain-specialized language models that connect domain knowledge with internal workflows in sensitive environments
A concise view of the jobs, capabilities and integrations described in the recorded product sources.
Documented product formats, platforms and official distribution destinations. Availability can vary by region and plan.
A concise history of software releases and material product changes. Events appear only when a dated source supports what changed.
Showing the newest updates and meaningful milestones. Open an entry for its summary and source.
Aleph Alpha published research introducing CHEETAH, an analytical model that simulates LLM inference speed from first principles by decomposing forward passes into compute-bound, memory-bound, and communication-bound GPU bottlenecks.
View source [7]Aleph Alpha published research proposing short soft-prompt tuning as an efficient method for fair LLM benchmark evaluation, separating model knowledge from formatting-following ability without a full post-training pipeline.
View source [3]Aleph Alpha research blog post describing a soft-prompt-tuning method that decouples a base LLM's factual knowledge from its ability to follow benchmark formatting, aimed at fairer evaluation of base models without full post-training.
View source [3]Aleph Alpha publishes a blog post on operational practices for ensuring consistent AI output quality in the public sector, drawing on lessons from the federal SPARK permitting acceleration project.
View source [4]Aleph Alpha blog post (Theresa Rickmann, Sven Büchel, Franca Ding) sharing lessons from the SPARK project for Germany's BMDS, presenting a framework for auditable AI output quality in regulated public-sector deployments.
View source [4]Facts, answers, structured details, milestones and primary resource links cite this shared ledger. Each external page appears once; release tags from the same GitHub project are grouped under one release history.
These tools share a workflow, capability or audience with Aleph Alpha. They may complement it rather than replace it, and are not presented as integrations or endorsements.
Savanna — Aleph Alpha's "model factory" that encodes the entire model training pipeline as code (Model Training as Code, MTaC)
Multi-phase training pipeline including pre-training, supervised fine-tuning (SFT), and reinforcement learning (RL)
KI Made in Germany / For Europe — Aleph Alpha
Aleph Alpha
Explainable, sovereign, and responsible AI
Critical enterprises and public institutions
Europe
Signed EU Code of Practice on Transparency of AI-generated Content
Signed EU's General-Purpose AI Code of Practice
Digital watermarks and clear provenance for AI-generated content
Open-source system for German federal AI infrastructure permitting
Real-time output evaluation using external Judge-LLMs (Prüf-Modelle)
AtMan (attention manipulation) explainability method, published at NeurIPS 2023
CHEETAH analytical model that simulates LLM inference speed from first principles
Aleph Alpha
Soft-Prompt-Tuning for fair and efficient evaluation of LLM benchmarks
Qwen3-32B (used as LLM-as-a-judge)
GSM8K
GSM8K dataset (math word problems)
Soft-Prompt-Tuning methodology (trains only ~10 soft-prompt vectors for ~80 steps on the benchmark training split)
company: Aleph Alpha · origin: Made in Germany, for Europe · audience: Enterprises and public institutions in Europe / critical enterprises and public
Aleph Alpha
Develops and operates customized, secure specialized language models (SLLMs) for business-critical environments
Enterprises and public institutions in Europe
Runs exclusively on European infrastructure
Models are trained in compliance with applicable EU law
Made in Germany, for Europe
Four German offices: Heidelberg (Speyerer Straße 14), Berlin (Wallstraße 16), Bayreuth (Weiherstraße 26), Munich (Einsteinstraße 174)
