TiRex - Best in Class Forecasting and Classifaction
Time Series are everywhere — in machinery, logistics, medicine, process industries, automotive systems, energy, and finance. While large language models address human language, our time series foundation model TiRex speaks the language of machines — the language of processes.
Compact
Compact
In-Context Learning
In-Context Learning
State Tracking
State Tracking
Try TiRex
We have uploaded a small TiRex demo to Huggingface. Anyone can try out the model there free of charge. We look forward to receiving your feedback.
Smaller is better – GIFT-Eval time-series leaderboard showing CRPS (Forecasting) rank, with NXAI’s TiRex as the top model
The idea behind TiRex
TiRex is an acronym for time and the Latin word for king, Rex. Our model is the king of time series. Because it works like the big LLMs. You feed a time series of your data into the model and TiRex delivers a forecast or classification – without any additional training.
The model is based on a new class of AI architectures: xLSTM. Despite having just 35 million parameters, TiRex is significantly smaller and more memory-efficient than its competitors. It not only excels in prediction accuracy but is also considerably faster. xLSTM time series models are designed to capture relationships over long time spans — a challenge that conventional transformer models often struggle with. In industrial applications, the reliable detection of such patterns is crucial, as they typically occur in subtle, hidden ways. Together with Google and Amazon, NXAI shares the top spots in the leaderboards.
TiRex excels existing time series models when it comes to classification tasks.
Classification is crucial
TiRex doesn't just excel at forecasting. It also outperforms other foundation models in the area of classification. Classification is often neglected in the evaluation of foundation models, but for industrial applications, classification is a very important factor.
Over 1,3 million downloads since June 25 can't be wrong. TiRex is available on Huggingface or as a Docker Container on GitHub.
Forecast, Classification and Domain-Specific TiRex models
TiRex is a foundation model for forecasting and classification. We currently use TiRex in mechanical engineering, logistics, and automation technology and robotics. TiRex runs on various devices, from PLCs to industrial PCs but also in an agentic setting. But forecasting and classification are only the first steps. Customers can use a domain-specific TiRex model to optimize their digital twin or develop and monetize customer-specific models more quickly.
Machinery manufacturers can offer customers TiRex-based solutions for optimization or commissioning. Thanks to in-context learning, the model adapts automatically to the customer's data – without retraining.
Explore our TiRex Research
TiRex for your application
Do you need support to deploy or finetune TiRex in your environment? We more than happy to help you.