/

Models Playground

FEV Bench

Point-forecast accuracy (MASE)

GIFT-Eval

Probabilistic and multiset robustness

BOOM

Infrastructure and observability forecasting

Impermanent

Temporal robustness and concept drift

ARFBench

Anomaly reasoning over observability telemetry

TIME

Task and pattern generalization

QuitoBench

Regime and context sensitivity

TSFM.ai Forecasting Playground

Run Chronos-Bolt and Mistral's Timestral entirely on your device. Upload a time series, forecast up to 64 steps, backtest against history, and inspect quantile forecasts. Inputs stay in your browser; model weights download from Hugging Face.

Loading playground...

Forecast a time series in your browser

Chronos-Bolt and Timestral run on this device, so your data stays here. Use a CSV with a time column and at least one numeric column. The models read up to 512 points and forecast up to 64 ahead (32 for Timestral).

Or start from a demo