€12m · Seed · AI · Heidelberg, Germany
Kausable, a Heidelberg-based startup developing causal “world models”, has raised €12m in a Seed round co-led by UVC Partners and Entourage, with HTGF and Mätch VC participating. According to UVC Partners, the company will use the capital to expand its then nine-person team and advance a frontier model designed to adapt to new conditions without repeated retraining.
The company was founded in 2025 by Johannes Haux, Dr Benjamin Herdeanu and Gregor Ramien after research at Heidelberg University. Its approach trains models on structural and causal priors, then asks them to update their view from limited new evidence. Kausable says that architecture is meant for dynamic environments where historical data is sparse or operating conditions change too quickly for a conventional training cycle.
From one-off model projects to reusable intelligence
The commercial wager is that adaptation can become part of inference rather than a new engineering project. Industrial AI deployments often require fresh data collection, labelling and retraining when equipment, markets or operating conditions shift. If Kausable can generalise its causal model across those changes, customers could reuse one intelligence layer across more systems instead of commissioning a separate model for every environment.
That would move the product closer to horizontal infrastructure, but the burden of proof is high. A system that claims to work across robotics, energy, finance and medicine must show reliable performance in very different domains, especially when the rare events it predicts are difficult to observe. The company has not announced commercial deployments, so the near-term milestone is technical validation rather than scaled distribution.
Research gives investors an early proof point
Kausable’s first named model, TipPFN, forecasts rare, high-impact transitions in complex dynamic systems. The company says it has tested the model across 14 domains and observed zero-shot performance; UVC Partners points to potential applications ranging from grid failures to medical events. Those are early research claims, but they illustrate why the investors are backing a foundational model rather than a single vertical application.
The founders previously raised about $1.5m in pre-seed funding in 2025. The new round finances a larger research effort and team, while the eventual business will depend on converting technical generality into repeatable customer value. For Kausable, avoiding retraining is not merely a model feature: it is the economic claim that could make ambitious AI projects cheaper to maintain as real-world conditions change.
Explore this theme
Judith Dada · 10 March 2026. The essay places world models inside the remaining frontier-AI opportunities where European teams may still build differentiated capability.
