Background: Co-leading the LLM effort of the Swiss AI Initiative alongside professors from EPFL, working on one of Europe's most ambitious sovereign AI projects. Schlag spearheads development of large-scale models on the ALPS supercomputer infrastructure with over 10,000 Grace Hopper superchips, representing Switzerland's comprehensive commitment to AI sovereignty. His work encompasses the full spectrum of AI development from foundational research to practical implementation, positioning Switzerland as a competitive force in global AI development while maintaining European values of transparency and responsibility.

The Swiss AI Initiative under his co-leadership represents a multi-faceted national commitment involving over 70 professors conducting generative AI research, with approximately 20 million Swiss francs committed over four years for PhDs and postdocs. The computational infrastructure is equally impressive, with 10 million GPU hours provided in 2024, scaling to 15-20 million hours annually thereafter. This represents one of the most significant sovereign AI investments outside the United States and China.

Key Perspective: Schlag brings crucial nuance to discussions of AI openness by distinguishing between "open weights" and truly "open source" AI models. His definition emphasizes transparency not just in model access but in the entire development process. « Open source means it is reproducible. So if you only share the weights of your model and you are not transparent about how you produce those weights, like what was the training process, and particularly also what training data do you use, then you cannot call yourself really open source. »

This distinction carries profound implications for regulatory compliance, community building, and scientific reproducibility. While open weights models allow usage and deployment, true open source enables collaborative improvement and verification—a difference that could determine the trajectory of AI development globally. Schlag argues that regulatory uncertainty makes this transparency increasingly valuable: « There is an increasing uncertainty when it comes to regulation in what is allowed and what not. And having an open source model is by nature much more transparent about what has been used to train these models. »

He provides concrete examples of how closed model licensing creates compliance complications for European businesses. « The Llama 4 models that came out—in the license, if you want to use that model, it explicitly says that you are not allowed to use the multimodal models in Europe. And basically all Llama 4 models are multimodal. » Such restrictions illustrate how dependence on proprietary systems can create unexpected business risks and regulatory complications.

Technical Leadership: Schlag outlines an ambitious technical approach that goes beyond simple fine-tuning to comprehensive model development from scratch. « We're training a model from scratch, currently focused on a pure text model, text in, text out. And then there's a pre-training stage which requires the most amount of compute, followed by a post-training stage. » The post-training phase encompasses conversational abilities, value alignment, and reasoning capabilities similar to those demonstrated by DeepSeek and OpenAI's o1 and o3 models.

His confidence in open source performance capabilities challenges conventional assumptions about proprietary model advantages. « Purely on performance, we are able, as in the open source communities, to increasingly close the gap between performance differences of very large models coming from these labs and open source efforts. » The Swiss initiative aims to demonstrate this principle by developing what he describes as « likely going to be the best performing model in this class since the open source class is still quite small when it comes to large language models. »

The Swiss initiative prioritizes inclusivity in language support, incorporating over 1,500 language and script pairs with particular attention to low-resource languages like Romansh. This multilingual focus reflects both Switzerland's linguistic diversity and a commitment to global accessibility that commercial models often neglect. The approach addresses a critical research question: how to improve performance on low-resource languages while maintaining competitive capabilities across major language groups.

Strategic Vision: Schlag emphasizes the strategic advantages of open source development for startup ecosystems and European competitiveness. « You need to build something that is maybe not just a wrapper around an existing service from which a US or Chinese tech company profits. If you can leverage this environment, this ecosystem, then you have a much easier time to find collaborators. » This ecosystem approach enables startups to influence model development directions rather than simply consuming predetermined capabilities.

He acknowledges the challenges of emerging technologies like multimodal capabilities while demonstrating thoughtful consideration of misuse potential versus utility. « If you have audio, audio output, general purpose audio output, are you sure that you can prevent that? The model could be easily used to clone voices, for example, to be fraudulent. And is this misuse potential actually in line with the potential benefit of the model speaking Swiss German? »

Schlag also addresses the infrastructure advantages Switzerland brings to AI development. The ALPS facility operates with over 12 years of CSCS experience in accelerated GPU-based supercomputers, knowledge that proves crucial for large-scale model training. « They're being trained right now on the Alps supercomputer, which has over 10,000 Grace Hopper superchips, GH200 GPUs which are basically the best GPUs you can get. » The environmental considerations add another dimension, with the facility operating primarily on hydroelectric power and implementing heat recovery systems for local district heating.

Future Implications: Schlag envisions AI systems that extend beyond simple text processing to integrated platforms combining language models with external tools and real-world interfaces. « What we will see in the future is that you can actually probably replace that engine which is the actual agent executing the actions and choosing which tools to use throughout different solutions. » This infrastructure perspective reinforces the strategic importance of open source approaches as foundational building blocks for next-generation technological systems.

Key Achievement: Imanol outlined Switzerland's comprehensive approach to sovereign AI development, positioning the nation to deliver what could be the best performing open source large language model while maintaining full transparency and multilingual inclusivity across 1,500+ language pairs. His leadership demonstrates how smaller nations can compete effectively with technological superpowers through strategic focus on openness, collaboration, and alignment with democratic values, while building sustainable competitive advantages through regulatory leadership and technical excellence.