Datadog has unveiled Toto 2, an open-source family of time series forecasting models, featuring versions with up to 2.5 billion parameters. Toto 2 is the first in its domain to validate the scaling law, showing improved predictive performance with increased parameters without saturation. The model family, released under the Apache 2.0 license, includes five sizes: 4M, 22M, 313M, 1B, and 2.5B. Toto 2 excels in evaluations across major forecasting benchmarks, ranking first in BOOM, GIFT-Eval, and TIME. It introduces a continuous patch masking mechanism, enhancing inference speed by replacing autoregressive generation with single-pass forward prediction. Notably, the 313M version matches the latency of Chronos-2, a smaller model. Toto 2 also demonstrates strong cross-domain generalization, outperforming competitors despite using only system monitoring metrics and synthetic data for pretraining.