LIVE-BENCH·last run 12:36 UTC·next 12:00·0/0 models·3/3 datasets healthywindow: last 7d · context: 512 · horizon: 24h
live-bench — every hosted TSFM, every hour, on data they couldn't have seen
Each hour we hand every hosted forecaster a 512-step context cut off 24 hours ago, ask for a 24-hour forecast, and score it against the realized values for that span. No data leakage by construction — the cutoff postdates every model's training.
runs scored · 7d0
inference calls · 24h0
caiso_lmp · TH_ZP26_GEN-APNDCAISO · LMP $/MWh · single node (post #824) · live
coinbase_prices · DOT/USDCoinbase · DOT-USD · last close per hour (post #824) · live
themeparks_wait_timesqueue-times.com · mean wait minutes across operating rides · live
rolling rank · last 7 days · compositelines flow into the model row on the right · top-5 highlighted
| # | model | score | Δ24h | Δ7d | streak |
|---|