This Online hackathon is an invitation to spend just over two weeks building with Apertus, Switzerland's fully open LLM. From 1–16 October, builders everywhere gather online to push Apertus 1.5 further. And, you're invited.

If you care about open and trustworthy AI that anyone can use, this is your place.

You'll work with both the 8B and 70B models, backed by our partners CSCS and Phoeniqs, and experienced mentors alongside you. Everything you ship stays open and reusable.
The strongest projects advances to the Grand Finals in May 2027 in St. Gallen, Switzerland where you'll compete for the Main «Hack Apertus» Award.

🎯🎯 TRACKS & CHALLENGES

🎯 TRACK 1 - APERTUS READINESS is about improving the model.

👉🏼 1 A: Red-Teaming Apertus
This challenge accepts submissions that find and document where Apertus underperforms, behaves unexpectedly, or produces outputs that do not meet the expectations of a sovereign, multilingual, trustworthy model. Participants can submit up to 5 issues. Any of the following topics, in any modality, are of particular interest in this hackathon:
· Misrepresentation of culture and values
· Stereotyping and bias
· Disclosure of private and personal information
· Generation of IP and copyrighted materials
· Misrepresentation of history and factual incorrectness

👉🏼 1 B: Swiss Voices
Localised alignment to the languages, dialects, knowledge and values of Switzerland

This challenge accepts submissions that evaluate any part of the lifecycle of localising Apertus to specific Swiss contexts, namely input adaptation, domain-specific core task intelligence, and behavioral steering. The lifecycle stages are motivated as follows.

Input adaptation to local speech and dialects:
Many speech-native multimodal LLMs struggle with phonetic drift, code-switching, local accents, and unwritten or low-resource dialects. Submissions for this lifecycle stage will measure phonetic transcription accuracy and dialect comprehension to evaluate Apertus’s ability to transcribe Swiss voice inputs into standard text representations without losing local context.

Core task intelligence for domain-specific knowledge:
General base models usually struggle with domain-specific knowledge and niche terminology. Submissions for this lifecycle stage will measure factuality and hallucination rates to evaluate whether Apertus possesses factually accurate, domain-tailored knowledge to execute specialised reasoning for localised Swiss contexts(e.g., legal statutes, local medical guidelines, enterprise workflows).

Behavioural steering for localised preferences:
Model performance is not only about dialects and facts, but also about cultural nuances, tone, safety boundaries, and local ethics. Submissions for this lifecycle stage will measure how frequently responses meet expectations to evaluate which regional norms and local regulatory standards Apertus respects by design.

🎯TRACK 2 - APERTUS ADOPTION is about showing what the model can do. 

👉🏼 2 A: Academia Challenges
Pick a challenge from our partners and solve it by building a prototype based on an Apertus model.

👉🏼 2 B: Own Project
Bring your own idea and build a working Apertus prototype that tackles a problem you care about — any domain, any use case. The project must be new, started within the hackathon period.

📅  FIRST STEPS & TIMELINE

a) Join our Discord, our main comms channel
b) Find team mates on Discord if you don't work solo (1-5 allowed)
c) 1. OCT, 12:00 CEST: Tracks and Challenges are released
d) Book a session with one of our mentors
e) Check for Online sessions to support you
f) Work remotely on your solution
g) 16. OCT, 12:00 CEST, Final submission deadline
h) Our expert jury reviews all submissions and selects the winners until 23 OCT 2026 

Note: This is the opening stage of a multi-month Hackathon journey. Over the following months, 6+ on-site hackathons across Switzerland pick up where the Online Hackathon leaves off, carrying your work forward with the on-site teams. Join one Hack or all of them. Online and On-site run as separate phases, so you choose how far you go.
The Winners from each Hack will compete for 🏆 Main «Hack Apertus» Award on 14.05.2027 in St.Gallen.

👥 Who Should Participate?

Open to all, recommended for
AI/ML engineers, software developers, data scientists, researchers, and NLP practitioners, plus technical students (CS, data science, computational linguistics) and non-engineer domain experts (law, medicine, dialects, public sector).
Bring a team or find one on Discord.

👉🏼 Choose your track, find your team, and join a growing community advancing Apertus and showing where it can go next. Wherever you start, you'll help shape a sovereign, trustworthy foundation others can rely and build on.

New to Apertus? It's an open large language model built by teams from EPFL, ETH Zurich and CSCS and released fully in the open, trained across 1,811 languages. Free to use, free to build on.

• Will be published soon.