What changed
IBM Research announced Granite Time Series PatchTST-FM-r2 on 9 September. The release provides model weights, inference code and reproducible benchmark materials. It supports forecasting without fitting a separate model to each dataset, alongside probabilistic outputs and missing-value imputation.
IBM reports strong GIFT-Eval results and offers Apache 2.0 or OpenMDW 1.0 licensing. Those performance claims come from the model’s publisher, not an independent evaluation by this site.
Why it matters
Business AI is broader than chat. Demand, traffic and operational measurements can be useful forecasting targets. Open materials make it easier to inspect and test a candidate before choosing a deployment approach.
Our view: compare forecasts with a simple baseline on data the model has not seen. Examine large misses as well as average accuracy, and keep uncertainty visible to the people making decisions.
Sources & further reading
Sources checked 2026-09-11. Product access and features may change. Practical guidance is our editorial interpretation.
