Datadog Toto 2.0: First Time Series Foundation Model with Reliable Scaling Laws

Datadog Toto 2.0: First Time Series Foundation Model with Reliable Scaling Laws

Datadog released Toto 2.0, a family of open-weights time series foundation models ranging from 4M to 2.5B parameters under Apache 2.0 license on HuggingFace. Every model size in the family beats the prior one from a single hyperparameter configuration—and Toto 2.0 is the first TSFM family where scaling laws produce a predictable, reproducible compute/data/performance curve. It leads the BOOM, GIFT-Eval, and TIME benchmarks simultaneously.

Why It Matters

Reliable scaling laws in time series models mean practitioners can now predict model performance at any compute budget before training—removing the trial-and-error cost that has made time series ML engineering expensive. Open Apache 2.0 licensing with full weights access enables on-premises deployment without API dependency.