Birla AI Labs launches Yug-1 to forecast demand and spot patterns in business data

Birla AI Labs has launched Yug-1, a 271.8-million-parameter AI model to forecast outcomes and identify unusual patterns in data that changes over time. The model’s potential uses lie across demand planning, equipment monitoring, finance and infrastructure.

The model, available on GitHub and Hugging Face, is the company's first time series foundation model (TSFM). It targets business information such as demand and supply trends, equipment performance and financial signals, which companies use to make decisions about their operations.

“Most of the conversation around AI today is about language. But so much of the data that actually runs businesses and economies moves through time: demand and supply, equipment performance, financial signals, infrastructure use,” Ananya Birla, founder and chairperson of Birla AI Labs, said in a LinkedIn post.


“Understanding how these patterns evolve is where some of the most consequential decisions get made,” she added.

Time series data consists of observations recorded over time, allowing users to examine how a measure changes and whether patterns repeat. For a business, the task could involve estimating how much demand it might face in the coming weeks or identifying an unexpected change in equipment performance.

Yug-1 uses what its technical documentation calls zero-shot probabilistic forecasting. It also provides a range of possible outcomes, allowing users to see the uncertainty around a prediction.That approach could let businesses test the same model on different forecasting problems without first developing a separate model for each one.

Birla said the model can forecast outcomes, find patterns and detect anomalies across complex, interconnected variables. The company has also positioned Yug-1 against other forecasting models, citing its performance on the GiftEval benchmark.

“On the GiftEval benchmark, it has outperformed other first-generation TSFMs and holds its own against newer-generation models. And our Frontier Research team took it from training to deployment in under six months,” she said.

The benchmark chart shared by Birla shows Yug recording 0.744 on a measure of forecasting error called MASE, where lower scores indicate better results. This puts it ahead of Google’s TimesFM 2.0 at 0.758 and Amazon’s Chronos Bolt Base at 0.808 on that measure, but behind IBM’s TTM-R3-PT at 0.724 and Salesforce’s Moirai2 at 0.728.

Birla also described Yug-1 as Asia’s first TSFM built for real-world use.

“When we started the research arm of Birla AI Labs, the belief was simple: research only matters when it reaches the real world,” she said.

The AI lab presented the work at NeurIPS 2025, ICML 2026 and the World Summit AI, as well as university campuses across India, Europe and the UK, Birla said.

It plans to return to NeurIPS 2026 and visit campuses and labs globally in the coming months to bring the model to technology and business leaders, she added.