FEV Bench

FEV Bench is a realistic benchmark for zero-shot time-series forecasting models across broad real-world forecasting tasks.

15rows
mase_skill_scoreprimary metric
2026-05-06sampled

Metadata

ID
fev_bench
Category
Data
Release
2025-09-30

Metrics

MASE Skill Score, MASE Win Rate, WQL Skill Score, WQL Win Rate, SQL Skill Score, SQL Win Rate, Median Training Time / 100 (lower is better), Median Inference Time / 100 (lower is better), Training Corpus Overlap (lower is better), Failures (lower is better)

Latest Results

Rows are ranked by MASE skill_score from the official full leaderboard CSV. WQL and SQL leaderboard metrics are joined by source model_name.

Rank Subject MASE Skill Score Model Match Provenance Sampled
1 Chronos-2 35.50 — Imported 2026-05-06
2 TimesFM-2.5 30.20 — Imported 2026-05-06
3 TiRex 30.01 — Imported 2026-05-06
4 Toto-1.0 28.21 — Imported 2026-05-06
5 TabPFN-TS 27.65 — Imported 2026-05-06
6 Moirai-2.0 27.26 — Imported 2026-05-06
7 Chronos-Bolt 26.52 — Imported 2026-05-06
8 Sundial-Base 24.75 — Imported 2026-05-06
9 Stat. Ensemble 15.65 — Imported 2026-05-06
10 AutoARIMA 11.24 — Imported 2026-05-06
11 AutoTheta 10.99 — Imported 2026-05-06
12 AutoETS 2.26 — Imported 2026-05-06
13 Seasonal Naive 0 — Imported 2026-05-06
14 Naive -16.67 — Imported 2026-05-06
15 Drift -18.14 — Imported 2026-05-06