TabPFN-TS
Treats forecasting as tabular regression: lightweight temporal features (lags, calendar) are fed into the pretrained tabular foundation model TabPFN-v2. No time-series-specific pretraining, yet SOTA on covariate-informed forecasting in GIFT-Eval at only 11M parameters.
TabPFN-TS is an unusual entry in this list: it isn't a time-series foundation model in the strict sense. Instead, it repurposes TabPFN-v2, a Prior-Fitted Network pretrained on purely tabular regression tasks, by adding a thin temporal featuriser (lags, calendar features, simple statistics) and feeding the result to TabPFN as a standard tabular row.
Despite skipping time-series-specific pretraining entirely, the 11M-parameter pipeline achieves state-of-the-art covariate-informed forecasting on GIFT-Eval and competitive univariate performance on fev-bench. It is also the smallest model in the lineup by an order of magnitude, which makes it cheap to deploy as a strong covariate baseline.
Versions on TS-Arena
Each version below corresponds to one registered model id in the leaderboard. Click through to its detail page for per-model rankings, forecasts, and history.
- TabPFN-TStabpfn-ts11M params…
Built on the TabPFN-v2 tabular foundation model; ultra-compact.