Toto 2.0: Time Series Forecasting Enters the Scaling Era

TL;DR

Toto 2.0 demonstrates that time series foundation models improve forecast quality with scale, releasing open models that set new benchmarks across multiple datasets.

Contribution

The paper introduces Toto 2.0, a scalable family of time series models trained with a unified recipe, achieving state-of-the-art results and providing open access to all checkpoints.

Findings

Models scale reliably from 4M to 2.5B parameters.

Toto 2.0 achieves new state-of-the-art on BOOM, GIFT-Eval, and TIME benchmarks.

Open-source release of all five base checkpoints.

Abstract

We show that time series foundation models scale: a single training recipe produces reliable forecast-quality improvements from 4M to 2.5B parameters. We release Toto 2.0, a family of five open-weights forecasting models trained under this recipe. The Toto 2.0 family sets a new state of the art on three forecasting benchmarks: BOOM, our observability benchmark; GIFT-Eval, the standard general-purpose benchmark; and the recent contamination-resistant TIME benchmark. This report describes our experimental results and details the design decisions behind Toto 2.0: its architecture and training recipe, training data, and the u-muP hyperparameter transfer pipeline. All five base checkpoints are released under Apache 2.0.