Datadog Launches Toto 2.0: A Breakthrough Time Series Forecasting Model Family That Scales to 2.5 Billion Parameters

Datadog launches Toto 2.0, a groundbreaking family of open-weights time series forecasting models scaling up to 2.5 billion parameters, achieving state-of-the-art results across all major benchmarks while being up to 7 times more parameter-efficient than its predecessor.

• Datadog releases Toto 2.0, a family of open-weights time series forecasting models ranging from 4 million to 2.5 billion parameters, marking the first TSFM family to demonstrate consistent, monotonic performance improvements with scale and no signs of saturation.
• Toto 2.0 achieves state-of-the-art results across all major benchmarks — BOOM, GIFT-Eval, and TIME — outperforming all competing foundation models, while also being up to 7 times more parameter-efficient than its predecessor and significantly faster at inference thanks to a new contiguous patch masking technique.
• Datadog identifies key frontiers for future TSFM development, including closing performance gaps with classical statistical baselines on long-horizon forecasting, improving data curation practices, treating observability metrics as a distinct data modality, and building full multimodal world models capable of reasoning across metrics, logs, traces, and events.