Codesota · Models · TimesFMGoogle Research8 results · 4 benchmarks
Model card

TimesFM.

Google Researchopen-source200M paramsDecoder-only Transformer

Zero-shot time-series foundation model. Pretrained on 100B time points.

§ 01 · Benchmarks

Every benchmark TimesFM has a recorded score for.

#BenchmarkArea · TaskMetricValueRankDateSource
01ETTm2Time Series · Time Series Forecastingmae0.3%#1/32025-02-02source ↗
02ETTm2Time Series · Time Series Forecastingmse0.3%#1/32025-02-02source ↗
03ETTh1Time Series · Time Series Forecastingmse0.5%#2/32025-02-02source ↗
04ETTh2Time Series · Time Series Forecastingmae0.4%#2/32025-02-02source ↗
05ETTh2Time Series · Time Series Forecastingmse0.4%#2/32025-02-02source ↗
06ETTm1Time Series · Time Series Forecastingmae0.4%#2/32025-02-02source ↗
07ETTm1Time Series · Time Series Forecastingmse0.4%#2/32025-02-02source ↗
08ETTh1Time Series · Time Series Forecastingmae0.4%#3/32025-02-02source ↗
Rank column shows this model’s position vs all other models scored on the same benchmark + metric (competitors after the slash). #1 in red means current SOTA. Sorted by rank, then newest result.
§ 02 · Strengths by area

Where TimesFM actually performs.

Time Series
4
benchmarks
avg rank #1.9
§ 03 · Papers

1 paper with results for TimesFM.

  1. 2025-02-02· Time Series· 8 results

    Sundial: A Family of Highly Capable Time Series Foundation Models

    Siqi Liu, Yujing Wang, Yuyao Zhang
§ 05 · Sources & freshness

Where these numbers come from.

Sundial Table 1
8
results
8 of 8 rows marked verified.