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.ai refetches the upstream Salesforce GIFT-Eval results every 12 hours via Next.js ISR. The 'last refreshed' timestamp at the bottom of the leaderboard reflects the most recent successful refresh.

Live·104 models · 104 evaluated·Auto-refreshed from the official leaderboard every 12 hours

GIFT-Eval Leaderboard

GIFT-Eval expands the question beyond raw point accuracy. It looks at how models perform across many datasets, frequencies, and forecasting settings, making it a strong benchmark for teams that care about robustness rather than a single flattering leaderboard slice. This page mirrors the official Salesforce GIFT-Eval leaderboard with search, filters, and per-slice rankings — covering every submitted model, not just the top few — and links each result back to a hosted endpoint on TSFM.ai where available.

CastStar ranks first with an average MASE rank of 16.32 across the GIFT-Eval dataset surface.

What this benchmark answers

Which models stay strong across heterogeneous datasets and probabilistic settings?

Methodology

Models are scored on grouped benchmark slices and ranked by average rank, with Weighted Quantile Loss providing a secondary read on probabilistic accuracy.

Showing 104 of 104 models · 38 hosted on TSFM.ai · Sorted by overall MASE rank (lower is better)

#	Model	Type	MASE rank	CRPS rank	MASE	CRPS	Leak	Repl. code	Hosted
1	CastStarHFCode USTC-AGI	Agentic	16.32	16.76	0.67	0.46	No	Yes	—
2	Cobra-AgentHF Dalpha AI	Agentic	17.68	17.15	0.68	0.46	No	No	—
3	PrismHF Birla AI Labs	Agentic	19.36	19.11	0.68	0.47	No	No	—
4	Toto-2.0-FnFHFCode Datadog	Agentic	19.38	19.08	0.68	0.46	No	Yes	—
5	RAES-Conductance-EnsembleHFCode California State University Northridge	Agentic	20.00	20.11	0.66	0.46	Yes	Yes	—
6	Taichu-TimeSeries-AgentHFCode zidongtaichu	Agentic	20.90	22.18	0.67	0.46	No	No	—
7	metis-autocastHF Metis	Agentic	21.35	18.29	0.68	0.46	No	No	—
8	Toto-2.0-2.5B-FTHFCode Datadog	Fine-tuned	21.99	20.86	0.68	0.46	No	Yes	—
9	TSOrchestraHFCode Melady Lab @ USC	Agentic	22.06	20.74	0.68	0.47	No	No	—
10	DeOSAlphaTimeGPTPredictor-2025HF vencortex®	Zero-shot	22.91	21.35	0.68	0.47	No	No	—
11	TimeRouterHFCode UConn & Morgan Stanley	Agentic	24.07	27.12	0.67	0.47	No	Yes	—
12	RAES-Conductance-Ensemble-VHFCode SETI	Agentic	24.12	24.44	0.67	0.46	No	Yes	—
13	MoiraiAgent-leakingHFCode Salesforce AI Research	Agentic	24.33	27.62	0.68	0.47	Yes	Yes	—
14	MoiraiAgentHFCode Salesforce AI Research	Agentic	26.37	28.73	0.69	0.48	No	Yes	—
15	CredenceHFCode ContinualIST	Agentic	26.47	23.89	0.69	0.47	No	Yes	—
16	Falcon-2.0HFCode ant-intl	Pretrained	28.33	37.02	0.67	0.49	No	Yes	—
17	Samay Kairosity	Agentic	29.10	26.24	0.70	0.48	No	No	—
18	TiRex-2-PretrainedHFCode NXAI	Pretrained	29.61	28.01	0.68	0.47	No	Yes	—
19	STRIDE (+Chronos-2)HF Google Cloud AI Research	Pretrained	29.90	24.88	0.67	0.45	No	No	—
20	Toto-2.0-2.5BHFCode Datadog	Pretrained	30.22	29.75	0.70	0.48	No	Yes	View model
21	Migas-1.0HF Synthefy	Pretrained	30.31	74.03	0.70	0.60	No	No	—
22	ZooCast-Top1Code	Agentic	30.32	30.98	0.69	0.48	No	No	—
23	TimeCopilotHFCode	Agentic	30.97	29.03	0.71	0.48	No	Yes	—
24	SynapseHF Google Cloud AI Research	Agentic	31.12	27.41	0.70	0.48	No	No	—
25	STRIDE (+Timer-S1)HF Google Cloud AI Research	Pretrained	31.14	29.06	0.67	0.46	No	No	—
26	Toto-2.0-1BHFCode Datadog	Pretrained	31.73	30.78	0.70	0.48	No	Yes	View model
27	Toto-2.0-313mHFCode Datadog	Pretrained	32.85	31.26	0.70	0.48	No	Yes	View model
28	Chronos-2HFCode AWS	Pretrained	33.35	33.79	0.70	0.49	No	Yes	View model
29	TurkForecast-FM-Chronos2-LoRA-v1HFCode TurkForecast (Mert Karatay)	Fine-tuned	34.77	34.14	0.66	0.42	No	Yes	—
30	TSOrchestra-testHF Melady Lab @ USC	Fine-tuned	35.44	26.24	0.70	0.47	Yes	No	—
31	TiRex-2-ZeroshotHFCode NXAI	Zero-shot	35.82	31.22	0.70	0.48	No	Yes	—
32	PatchTST-FM-r1HFCode IBM TSFM & Rensselaer Polytechnic Institute	Zero-shot	36.61	33.29	0.71	0.48	No	Yes	View model
33	Timer-S1HFCode Tsinghua & ByteDance	Pretrained	37.09	37.58	0.69	0.49	No	Yes	View model
34	ValBestSingle-cmttHFCode G-connor	Agentic	37.14	75.44	0.71	0.60	No	No	—
35	Granite-FlowState-r1.1HFCode IBM TSFM	Zero-shot	37.16	37.65	0.70	0.49	No	Yes	—
36	TimesFM-2.5HFCode Google Research	Zero-shot	37.74	36.69	0.71	0.49	No	Yes	View model
37	FlowState-r1.1HFCode IBM TSFM	Zero-shot	37.97	36.40	0.70	0.49	No	Yes	View model
38	LongSeer-v1.0 LongShine AI Research	Zero-shot	39.59	39.73	0.71	0.49	No	No	—
39	Falcon-XHFCode ant-intl	Pretrained	41.03	40.36	0.69	0.49	No	Yes	—
40	TiRexHFCode NXAI	Zero-shot	41.45	34.89	0.72	0.49	No	Yes	View model
41	Granite-PatchTST-FM-r1HFCode IBM TSFM & Rensselaer Polytechnic Institute	Zero-shot	42.03	36.62	0.72	0.49	No	Yes	View model
42	Reverso MIT	Zero-shot	42.36	76.26	0.71	0.61	No	No	—
43	Toto-2.0-22mHFCode Datadog	Pretrained	42.70	38.08	0.72	0.50	No	Yes	View model
44	Chronos-2-SynthHFCode AWS	Zero-shot	43.55	42.14	0.72	0.50	No	Yes	View model
45	Xihe-ultraHF Ant	Zero-shot	43.63	41.36	0.70	0.49	No	No	—
46	TTM-R3-FTHFCode IBM TSFM	Fine-tuned	44.33	46.99	0.71	0.51	No	Yes	View model
47	VISIT-2.0	Pretrained	44.87	41.88	0.71	0.50	No	No	—
48	t0-alphaHFCode The Forecasting Company	Pretrained	45.85	37.35	0.72	0.49	No	Yes	—
49	Xihe-maxHF Ant	Zero-shot	46.69	43.55	0.71	0.49	No	No	—
50	TEMPO_ENSEMBLEHF Melady Lab @ USC	Fine-tuned	47.57	39.55	0.79	0.46	Yes	No	—
51	TTM-R3-PTHFCode IBM TSFM	Pretrained	48.03	51.42	0.72	0.52	No	Yes	View model
52	Reverso-SmallHFCode MIT	Zero-shot	48.40	78.76	0.73	0.63	No	Yes	—
53	FlowState-9.1MHFCode IBM Research	Zero-shot	48.43	42.47	0.73	0.50	No	Yes	—
54	A Kairos_50mHFCode ShanghaiTech University	Zero-shot	50.30	61.20	0.74	0.55	No	Yes	View model
55	Moirai2HFCode Salesforce AI Research	Pretrained	51.90	49.38	0.73	0.52	No	Yes	View family
56	Toto_Open_Base_1.0HFCode Datadog	Zero-shot	53.81	47.94	0.75	0.52	No	Yes	View model
57	A Kairos_23mHFCode ShanghaiTech University	Zero-shot	54.40	62.95	0.75	0.55	No	Yes	View model
58	A Kairos_10mHFCode ShanghaiTech University	Zero-shot	54.99	64.08	0.75	0.55	No	Yes	View model
59	chronos_bolt_baseHFCode AWS AI Labs	Pretrained	55.04	56.06	0.81	0.57	Yes	Yes	View model
60	TTM-R2-FinetunedHFCode IBM Research	Fine-tuned	55.05	65.24	0.76	0.58	Yes	Yes	View model
61	CHARMHFCode C3 AI	Zero-shot	55.34	24.74	0.76	0.48	No	No	—
62	Toto-2.0-4mHFCode Datadog	Pretrained	55.49	50.94	0.76	0.52	No	Yes	View model
63	timesfm_2_0_500mHFCode Google Research	Pretrained	56.68	55.79	0.76	0.55	Yes	Yes	View model
64	sundial_base_128mHFCode Tsinghua University	Zero-shot	58.70	63.42	0.75	0.56	No	Yes	View model
65	xLSTM-MixerHF AIML Lab @ TU Darmstadt	Deep learning	60.42	46.99	0.78	0.51	No	No	—
66	Reverso-Nano MIT	Zero-shot	61.27	82.51	0.76	0.66	No	No	—
67	CleanTS-65MHFCode Shandong University	Zero-shot	61.33	53.01	0.80	0.54	No	Yes	—
68	chronos_bolt_smallHFCode AWS AI Labs	Pretrained	61.82	60.11	0.82	0.58	Yes	Yes	View model
69	PatchFMHFCode LITIS	Zero-shot	63.42	57.14	0.81	0.56	No	Yes	—
70	YingLong_300mHFCode Alibaba	Zero-shot	63.67	56.94	0.80	0.55	No	Yes	View model
71	YingLong_110mHFCode Alibaba	Zero-shot	64.44	58.07	0.81	0.56	No	Yes	View model
72	TabPFN-TSHFCode PriorLabs	Zero-shot	64.65	57.67	0.77	0.54	No	Yes	—
73	TempoPFNHFCode University of Freiburg	Zero-shot	64.77	55.13	0.79	0.53	No	Yes	—
74	YingLong_50mHFCode Alibaba	Zero-shot	67.57	61.19	0.82	0.57	No	Yes	View model
75	Lingjiang Alibaba Cloud	Zero-shot	68.61	63.77	1.00	0.62	No	No	—
76	Moirai_largeHFCode Salesforce AI Research	Zero-shot	68.96	61.92	0.87	0.60	No	Yes	View model
77	Moirai_baseHFCode Salesforce AI Research	Zero-shot	69.72	61.91	0.90	0.61	No	Yes	View model
78	Chronos_largeHFCode AWS AI Labs	Pretrained	72.03	74.29	0.87	0.65	Yes	Yes	View model
79	Chronos_baseHFCode AWS AI Labs	Pretrained	72.51	74.92	0.88	0.65	Yes	Yes	—
80	PatchTST Princeton University	Deep learning	75.06	67.43	0.85	0.59	No	No	View model
81	FLAIRHFCode Mellon Inc.	Statistical	75.29	65.88	0.84	0.59	No	Yes	—
82	Chronos_smallHFCode AWS AI Labs	Pretrained	75.61	76.02	0.89	0.66	Yes	Yes	—
83	i_transformer Tsinghua University	Deep learning	76.84	69.69	0.89	0.62	No	Yes	—
84	YingLong_6mHFCode Alibaba	Zero-shot	77.64	70.74	0.88	0.61	No	Yes	View model
85	Super-LinearHFCode Ben-Gurion University of the Negev	Pretrained	77.82	89.11	0.86	0.74	Yes	Yes	—
86	TimesFMHFCode Google Research	Pretrained	79.54	74.14	1.08	0.68	Yes	Yes	View family
87	TFT Google Research	Deep learning	79.74	67.94	0.92	0.60	No	No	—
88	VisionTSHF Zhejiang University	Zero-shot	81.45	89.84	0.86	0.75	No	No	—
89	Moirai_smallHFCode Salesforce AI Research	Zero-shot	81.58	72.39	0.95	0.65	No	Yes	View model
90	N-BEATS ServiceNow	Deep learning	82.39	90.52	0.94	0.82	No	No	—
91	FFMHFCode IBM Research	Deep learning	82.87	76.71	1.02	0.70	No	No	—
92	Auto_Arima	Statistical	84.22	89.10	1.07	0.91	No	No	—
93	TTM-R2-PretrainedHFCode IBM Research	Pretrained	85.38	93.61	1.02	0.87	Yes	Yes	View model
94	TTM-R1-PretrainedHFCode IBM Research	Pretrained	87.35	93.82	1.08	0.89	Yes	Yes	View model
95	Auto_ETS	Statistical	87.57	92.40	1.21	7.49	No	No	—
96	Auto_Theta	Statistical	87.74	89.99	1.09	1.24	No	No	—
97	Seasonal_NaiveHFCode	Statistical	88.18	93.57	1.00	1.00	No	Yes	—
98	Crossformer Shanghai Jiao Tong University	Deep learning	88.77	80.21	2.57	1.64	No	No	—
99	DLinear The Chinese University of Hong Kong	Deep learning	89.13	93.57	1.06	0.85	No	No	—
100	DeepAR Amazon Research	Deep learning	90.35	83.53	1.34	0.85	No	No	—
101	TIDE Google Research	Deep learning	90.36	85.04	1.09	0.77	No	No	—
102	Lag-LlamaHF Morgan Stanley & Service Now	Pretrained	93.65	90.66	1.23	0.88	Yes	No	View model
103	NaiveHFCode	Statistical	93.93	98.45	1.27	1.59	No	Yes	—
104	VISIT-1.0	Pretrained	94.10	98.58	1.27	1.59	No	No	—

104 models · Aggregated live from the official GIFT-Eval leaderboard. Refreshes every 12 hours.Last refreshed Jul 15, 2026, 2:37 PM

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: 29(28%)
Zero-shot: 35(34%)
Fine-tuned: 6(6%)
Agentic: 18(17%)
Deep learning: 10(10%)
Statistical: 6(6%)

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.ai refetches the upstream Salesforce GIFT-Eval results every 12 hours via Next.js ISR. The 'last refreshed' timestamp at the bottom of the leaderboard reflects the most recent successful refresh.

Title	Date
GIFT-Eval: Salesforce's Comprehensive TSFM Benchmark	2026-02-15	Read →
The State of Multivariate Forecasting in 2025	2025-05-08	Read →
The Challenges of Benchmarking TSFMs	2024-11-05	Read →