AI Forecasting Breakthrough: TiRex Sets New Global Standard with IT:U Research Involvement
A newly published scientific study has set a new benchmark in artificial intelligence forecasting — and IT:U is part of the story. The paper introduces TiRex, a compact and efficient AI model capable of predicting future data trends without prior task-specific training, a method known as zero-shot forecasting.
What is TiRex?
TiRex was developed by the Linz-based AI startup NXAI and leading researchers from prestigious institutions, including the LIT AI Lab at the JKU and the Interdisciplinary Transformation University Austria (IT:U).
The model sets a new global benchmark in AI forecasting and features IT:U founding professor Daniel Klotz — Professor of the Machine Learning in Earth Science research group — as one of its co-authors, reflecting the university’s strong foundation in interdisciplinary, high-impact research.
TiRex is a powerful new model designed to forecast time-series data with high accuracy and efficiency. Built on the advanced xLSTM architecture, it combines the long-term memory strength of LSTMs with the modern in-context learning capabilities of Transformers.
Competing With Giants — and Winning
With only 35 million parameters, TiRex is significantly smaller than models from Google (TimesFM), Amazon (Chronos Bolt), Salesforce (Moirai), and Prior Labs (TabPFN-TS). Yet it outperformed them all in standardized tests like GiftEval and Chronos-ZS, delivering higher accuracy and faster results. This makes TiRex a standout in zero-shot forecasting, where the goal is to apply a model to completely new data without retraining — a task that typically requires large, resource-heavy models.
How TiRex make the Real-World Impact
Beyond the benchmarks, TiRex has real-world relevance. Its zero-shot capabilities allow it to be applied to novel time series without retraining. Thanks to its efficient design and strong generalization ability, it’s ideally suited for industrial applications — from predicting energy consumption to optimizing supply chains or monitoring machine health. Its compact architecture (notably, the low number of parameters and test-time scaling enabled by xLSTM) allows it to outperform larger models in both speed and resource use.
Unlike many AI systems, TiRex can be embedded directly into hardware, enabling real-time predictions in environments with limited computational capacity. This opens up new possibilities in fields such as logistics, automotive systems, and robotics.
“Advancing AI means making machine learning more accessible and practical. In many industrial applications, we don’t have the data, resources, or expertise to retrain or adapt complex models. TiRex shows how we can lower those barriers — a goal that reflects what we’re building at IT:U.”
— Daniel Klotz, Founding Professor, IT:U
Interdisciplinary Expertise Behind TiRex
The TiRex project is the result of a cross-institutional collaboration between:
- Andreas Auer – NXAI GmbH (Linz-based AI startup) & ELLIS Unit, LIT AI Lab, Institute for Machine Learning, JKU Linz
- Patrick Podest – ELLIS Unit, LIT AI Lab, Institute for Machine Learning, JKU Linz
- Daniel Klotz – Interdisciplinary Transformation University Austria (IT:U)
- Sebastian Böck – NXAI GmbH
- Günter Klambauer – ELLIS Unit, LIT AI Lab, Institute for Machine Learning, JKU Linz
- Sepp Hochreiter – NXAI GmbH & LLIS Unit, LIT AI Lab, Institute for Machine Learning, JKU Linz
This partnership brings together the innovation of an AI startup with the research depth of academic institutions, with contributors like IT:U’s Professor Daniel Klotz helping to advance the kind of interdisciplinary, high-impact research that aligns with IT:U’s academic vision.