Impermanent

Impermanent is a live benchmark for temporal generalization in time-series forecasting, evaluating models sequentially as GitHub activity outcomes arrive.

18rows
average_maseprimary metric
2026-05-06sampled

Metadata

ID
impermanent
Category
Data
Release
2026-03-09

Metrics

Average MASE (lower is better), Average Scaled CRPS (lower is better), Average MASE Rank (lower is better), Average Scaled CRPS Rank (lower is better)

Latest Results

Rows are parsed from the embedded live leaderboard DATA object and ranked by average MASE, lower is better. The source summary rows are preserved without model canonicalization.

Rank Subject Average MASE Model Match Provenance Sampled
1 AutoNHITS 0.39 — Imported 2026-05-06
2 AutoNBEATS 0.39 — Imported 2026-05-06
3 TiRex 0.40 — Imported 2026-05-06
4 Chronos 0.40 — Imported 2026-05-06
5 AutoTFT 0.41 — Imported 2026-05-06
6 TimesFM 0.43 — Imported 2026-05-06
7 Moirai 0.45 — Imported 2026-05-06
8 AutoPatchTST 0.46 — Imported 2026-05-06
9 FlowState 0.51 — Imported 2026-05-06
10 ZeroModel 0.53 — Imported 2026-05-06
11 SeasonalNaive 0.65 — Imported 2026-05-06
12 AutoCES 0.68 — Imported 2026-05-06
13 DynamicOptimizedTheta 0.68 — Imported 2026-05-06
14 AutoETS 0.81 — Imported 2026-05-06
15 AutoARIMA 0.96 — Imported 2026-05-06
16 Toto 1.77 — Imported 2026-05-06
17 Prophet 1.89 — Imported 2026-05-06
18 HistoricAverage 3.09 — Imported 2026-05-06