GIFT-Eval Leaderboard

Model landscape

GIFT-Eval is not only a TSFM leaderboard — it includes classical statistical baselines (ARIMA, ETS, Theta), deep-learning architectures (PatchTST, iTransformer, TFT), and agentic systems, so foundation-model results can be compared against the full prior art.

Pretrained

44(28%)

Zero-shot

51(32%)

Fine-tuned

7(4%)

Agentic

39(25%)

Deep learning

11(7%)

Statistical

6(4%)

Why robustness across datasets matters

A model that tops one leaderboard can collapse on a different frequency or domain. GIFT-Eval forces models to prove themselves across 23 grouped dataset slices covering different frequencies, horizons, and series shapes. If a model ranks well here, you can be more confident it will not surprise you when your data does not look like the training distribution.

Understanding Weighted Quantile Loss

WQL penalizes both overconfident and underconfident prediction intervals. A model with a low WQL produces forecast distributions that are well-calibrated — the 90th percentile prediction actually lands above the true value about 90% of the time. This matters for capacity planning, inventory, and any decision that depends on reliable uncertainty estimates rather than just the median forecast.

How to interpret it

• —Lower Average Rank is better because the leaderboard aggregates placement across many slices.
• —WQL matters when forecast calibration and uncertainty quality matter to the business.
• —Use GIFT-Eval to sanity-check whether a model is robust beyond a single narrow domain.

Frequently asked questions

What is GIFT-Eval?

GIFT-Eval (General Time Series Forecasting Model Evaluation) is a probabilistic forecasting benchmark from Salesforce that evaluates time series foundation models across 23 diverse dataset groups spanning 7 domains and 10 frequencies. Models are ranked by Average Rank and Average Weighted Quantile Loss.

What does Average Rank mean in GIFT-Eval?

Average Rank is the mean placement of a model across every benchmark slice. A lower value means the model consistently finishes near the top across heterogeneous datasets rather than dominating only one slice.

When should I use GIFT-Eval instead of FEV Bench?

Use GIFT-Eval when you care about probabilistic forecast quality (calibrated uncertainty) and robustness across many different data domains, rather than just point-forecast accuracy on a single leaderboard.

Does GIFT-Eval test multivariate forecasting?

GIFT-Eval includes both univariate and multivariate slices. The leaderboard on TSFM.ai aggregates both variate types from the upstream grouped-by-univariate file, and you can filter per-slice rankings to inspect multivariate-only performance.

What is the 'test leak' column?

Some GIFT-Eval submissions are from models whose training data overlaps with the evaluation datasets. A 'Yes' in the test-leak column means the authors disclosed partial or full pretraining-data overlap; use the filter to compare only models with no known test-data leakage.

What model types appear on the leaderboard?

GIFT-Eval ranks pretrained foundation models, zero-shot models, fine-tuned models, agentic systems, classical deep-learning architectures (PatchTST, iTransformer, TFT, TiDE, N-BEATS), and statistical baselines (ARIMA, ETS, Theta). The page shows every type so you can benchmark a TSFM against classical prior art.

How often is the leaderboard refreshed?

TSFM caches the discovered per-model result files for up to 12 hours and revalidates them on visits. The displayed page-data load time is not an author publication timestamp. Failed source files are disclosed as partial coverage.

Related reading

Compare with other TSFM benchmarks