IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license

Time-series foundation models are changing how forecasting systems are built: instead of training and maintaining a separate model for every dataset, users can use a pretrained model to generate forecasts zero-shot. IBM has released Granite Time Series PatchTST-FM-r2, the latest model in the Granite TSFM family and a successor to PatchTST-FM-r1, combining an updated architecture, a larger pretraining corpus, probabilistic forecasting, support for imputing missing values, and strong zero-shot performance in a roughly 385M-parameter model.

PatchTST-FM-r2 is designed as a general-purpose zero-shot forecaster for demand, prices, energy loads, traffic, telemetry, and other time series. It supports context lengths up to 8,192, flexible forecast lengths, and probabilistic forecasts through a 99-quantile prediction head. Its backbone uses conformer blocks that combine multi-head self-attention with temporal convolution to capture long- and short-range temporal structure.

As of September 8, 2026, the model is the top-performing zero-shot model released under a permissive, commercial-friendly open-source license (Apache 2.0 and OpenMDW 1.0) among replicable zero-shot models on the GIFT-Eval leaderboard. GIFT-Eval is a comprehensive time-series forecasting benchmark designed to evaluate models across diverse forecasting scenarios, and PatchTST-FM-r2 ranks #2 overall among replicable zero-shot models. When the leaderboard is restricted to models that are zero-shot, replicable, and evaluated without test leakage, PatchTST-FM-r2 ranks second for both CRPS and MASE, where lower values are better.

Model weights, architecture, inference pipeline, and code needed to reproduce the benchmark results are all available. The accompanying blog describes the model, benchmarking results, architecture, training data, and licensing, provides code examples, and highlights how Granite Time Series family models can be used in streaming production settings with Confluent. The model is dual-licensed under Apache-2.0 and OpenMDW-1.0, with users allowed to select either license, and it is available on Hugging Face.