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·130 models · 130 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.

STRIDE w/ Synapse ranks first with an average MASE rank of 14.62 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 130 of 130 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	STRIDE w/ SynapseHF Google Cloud AI Research	Agentic	14.62	13.63	0.63	0.42	No	No	—
2	EXAONE-Forecast-Agent LG AI Research	Agentic	18.87	19.49	0.61	0.42	No	No	—
3	LS-MoE LongShine AI Research	Agentic	19.31	20.16	0.65	0.44	No	No	—
4	LS-Agent LongShine AI Research	Agentic	21.25	22.41	0.65	0.45	No	No	—
5	Falcon-Agent ant-intl	Agentic	22.15	23.19	0.67	0.45	No	No	—
6	CastStarHF USTC-AGI	Agentic	22.23	20.79	0.66	0.45	No	No	—
7	limix_moe	Agentic	22.24	25.01	0.66	0.45	No	No	—
8	TIMEHEDGE Neurogica Inc.	Agentic	23.66	22.80	0.65	0.44	No	No	—
9	RegLab-AH-1 RegLab	Agentic	24.31	25.01	0.66	0.45	No	No	—
10	RacineCast-1-NCHFCode Racine.ai	Agentic	25.64	25.91	0.67	0.46	No	Yes	—
11	RacineCast-1HFCode Racine.ai	Agentic	25.75	26.86	0.67	0.46	No	Yes	—
12	FastCastrage sme	Agentic	27.27	27.54	0.67	0.46	No	No	—
13	TimesFM-3HFCode Google Research	Zero-shot	27.70	25.99	0.67	0.46	No	Yes	—
14	Cobra-AgentHF Dalpha AI	Agentic	28.60	27.91	0.68	0.46	No	No	—
15	Toto-2.0-FnFHFCode Datadog	Agentic	29.41	29.55	0.68	0.46	No	Yes	—
16	RAES-Conductance-EnsembleHFCode California State University Northridge	Agentic	30.47	30.90	0.66	0.46	Yes	Yes	—
17	PrismHF Birla AI Labs	Agentic	31.19	30.80	0.68	0.47	No	No	—
18	EXAONE-Forecast LG AI Research	Zero-shot	31.61	29.74	0.67	0.46	No	No	—
19	HistRoute-CV Hitsz iLearn-ML	Agentic	32.00	32.67	0.68	0.47	No	No	—
20	Taichu-TimeSeries-AgentHFCode zidongtaichu	Agentic	32.52	34.39	0.67	0.46	No	No	—
21	metis-autocastHF Metis	Agentic	32.78	28.70	0.68	0.46	No	No	—
22	Toto-2.0-2.5B-FTHFCode Datadog	Fine-tuned	33.26	32.58	0.68	0.46	No	Yes	—
23	TSOrchestraHFCode Melady Lab @ USC	Agentic	33.55	31.96	0.68	0.47	No	No	—
24	DeOSAlphaTimeGPTPredictor-2025HF vencortex®	Zero-shot	34.54	33.15	0.68	0.47	No	No	—
25	ForecastMate Zhejiang University	Agentic	34.73	33.62	0.66	0.46	No	No	—
26	RAES-Conductance-Ensemble-VHFCode SETI	Agentic	35.95	36.25	0.67	0.46	No	Yes	—
27	TimeRouterHFCode UConn & Morgan Stanley	Agentic	36.25	40.51	0.67	0.47	No	Yes	—
28	MoiraiAgent-leakingHFCode Salesforce AI Research	Agentic	37.40	41.58	0.68	0.47	Yes	Yes	—
29	Falcon-2.0HFCode ant-intl	Pretrained	39.61	51.26	0.67	0.49	No	Yes	—
30	Granite-PatchTST-FM-r2HFCode IBM TSFM & Rensselaer Polytechnic Institute	Zero-shot	39.77	37.31	0.68	0.47	No	Yes	—
31	MoiraiAgentHFCode Salesforce AI Research	Agentic	40.08	43.07	0.69	0.48	No	Yes	—
32	CredenceHFCode ContinualIST	Agentic	40.14	37.26	0.69	0.47	No	Yes	—
33	STRIDE (+Chronos-2)HF Google Cloud AI Research	Pretrained	41.22	35.43	0.67	0.45	No	No	—
34	Samay Kairosity	Agentic	43.30	40.04	0.70	0.48	No	No	—
35	TiRex-2-PretrainedHFCode NXAI	Pretrained	43.72	42.25	0.68	0.47	No	Yes	—
36	Toto-2.0-2.5BHFCode Datadog	Pretrained	43.91	44.31	0.70	0.48	No	Yes	View model
37	STRIDE (+Timer-S1)HF Google Cloud AI Research	Pretrained	44.20	42.02	0.67	0.46	No	No	—
38	Migas-1.0HF Synthefy	Pretrained	44.68	97.65	0.70	0.60	No	No	—
39	ZooCast-Top1Code	Agentic	44.69	45.60	0.69	0.48	No	No	—
40	TimeCopilotHFCode	Agentic	45.89	43.60	0.71	0.48	No	Yes	—
41	SynapseHF Google Cloud AI Research	Agentic	45.99	41.60	0.70	0.48	No	No	—
42	Toto-2.0-1BHFCode Datadog	Pretrained	46.46	45.70	0.70	0.48	No	Yes	View model
43	Toto-2.0-313mHFCode Datadog	Pretrained	47.63	46.26	0.70	0.48	No	Yes	View model
44	TurkForecast-FM-Chronos2-LoRA-v1HFCode TurkForecast (Mert Karatay)	Fine-tuned	47.84	47.26	0.66	0.42	No	Yes	—
45	Chronos-2HFCode AWS	Pretrained	48.03	48.60	0.70	0.49	No	Yes	View model
46	EFG-base LAIR	Zero-shot	49.96	49.21	0.70	0.48	No	No	—
47	TSOrchestra-testHF Melady Lab @ USC	Fine-tuned	50.39	39.33	0.70	0.47	Yes	No	—
48	ZeusHFCode GestaltCog Lab @ ICT	Pretrained	51.15	45.96	0.69	0.48	No	Yes	—
49	tafsutHF Tafsut-FM (Huawei GTS x EURECOM)	Pretrained	51.16	50.12	0.69	0.48	No	Yes	—
50	TiRex-2-ZeroshotHFCode NXAI	Zero-shot	51.25	45.97	0.70	0.48	No	Yes	—
51	Granite-FlowState-r1.1HFCode IBM TSFM	Zero-shot	52.09	53.89	0.70	0.49	No	Yes	—
52	PatchTST-FM-r1HFCode IBM TSFM & Rensselaer Polytechnic Institute	Zero-shot	52.26	48.76	0.71	0.48	No	Yes	View model
53	Timer-S1HFCode Tsinghua & ByteDance	Pretrained	52.59	53.67	0.69	0.49	No	Yes	View model
54	ValBestSingle-cmttHFCode N/A	Agentic	52.94	99.18	0.71	0.60	No	No	—
55	FlowState-r1.1HFCode IBM TSFM	Zero-shot	53.05	52.25	0.70	0.49	No	Yes	View model
56	TimesFM-2.5HFCode Google Research	Zero-shot	53.12	52.69	0.71	0.49	No	Yes	View model
57	Falcon-XHFCode ant-intl	Pretrained	55.86	55.89	0.69	0.49	No	Yes	—
58	LongSeer-v1.0 LongShine AI Research	Zero-shot	56.20	56.63	0.71	0.49	No	No	—
59	TiRexHFCode NXAI	Zero-shot	57.84	50.68	0.72	0.49	No	Yes	View model
60	Granite-PatchTST-FM-r1HFCode IBM TSFM & Rensselaer Polytechnic Institute	Zero-shot	58.58	52.64	0.72	0.49	No	Yes	View model
61	Reverso MIT	Zero-shot	59.12	100.4	0.71	0.61	No	No	—
62	Toto-2.0-22mHFCode Datadog	Pretrained	59.53	54.33	0.72	0.50	No	Yes	View model
63	Xihe-ultraHF Ant	Zero-shot	60.11	57.93	0.70	0.49	No	No	—
64	Chronos-2-SynthHFCode AWS	Zero-shot	60.47	59.28	0.72	0.50	No	Yes	View model
65	TEMPO_ENSEMBLEHF Melady Lab @ USC	Fine-tuned	61.21	52.54	0.79	0.46	Yes	No	—
66	TTM-R3-FTHFCode IBM TSFM	Fine-tuned	61.60	65.63	0.71	0.51	No	Yes	View model
67	VISIT-2.0	Pretrained	62.02	59.13	0.71	0.50	No	No	—
68	t0-alphaHFCode The Forecasting Company	Pretrained	63.36	53.35	0.72	0.49	No	Yes	—
69	Xihe-maxHF Ant	Zero-shot	63.84	60.59	0.71	0.49	No	No	—
70	TTM-R3-PTHFCode IBM TSFM	Pretrained	65.88	70.67	0.72	0.52	No	Yes	View model
71	FlowState-9.1MHFCode IBM Research	Zero-shot	66.02	59.82	0.73	0.50	No	Yes	—
72	Reverso-SmallHFCode MIT	Zero-shot	66.54	103.4	0.73	0.63	No	Yes	—
73	A Kairos_50mHFCode ShanghaiTech University	Zero-shot	68.02	82.68	0.74	0.55	No	Yes	View model
74	Moirai2HFCode Salesforce AI Research	Pretrained	70.04	67.76	0.73	0.52	No	Yes	View family
75	Toto_Open_Base_1.0HFCode Datadog	Zero-shot	72.82	66.28	0.75	0.52	No	Yes	View model
76	A Kairos_23mHFCode ShanghaiTech University	Zero-shot	73.02	84.89	0.75	0.55	No	Yes	View model
77	chronos_bolt_baseHFCode AWS AI Labs	Pretrained	73.65	75.62	0.81	0.57	Yes	Yes	View model
78	A Kairos_10mHFCode ShanghaiTech University	Zero-shot	74.03	85.95	0.75	0.55	No	Yes	View model
79	TTM-R2-FinetunedHFCode IBM Research	Fine-tuned	74.24	87.48	0.76	0.58	Yes	Yes	View model
80	CHARMHFCode C3 AI	Zero-shot	74.78	35.52	0.76	0.48	No	No	—
81	timesfm_2_0_500mHFCode Google Research	Pretrained	74.84	74.96	0.76	0.55	Yes	Yes	View model
82	Toto-2.0-4mHFCode Datadog	Pretrained	74.95	69.92	0.76	0.52	No	Yes	View model
83	sundial_base_128mHFCode Tsinghua University	Zero-shot	77.97	85.36	0.75	0.56	No	Yes	View model
84	xLSTM-MixerHF AIML Lab @ TU Darmstadt	Deep learning	80.15	64.90	0.78	0.51	No	No	—
85	recursive-moirai-2HFCode Emilio Cantu	Pretrained	80.66	73.14	0.77	0.53	No	Yes	—
86	CleanTS-65MHFCode Shandong University	Zero-shot	81.36	72.47	0.80	0.54	No	Yes	—
87	chronos_bolt_smallHFCode AWS AI Labs	Pretrained	81.38	80.63	0.82	0.58	Yes	Yes	View model
88	Reverso-Nano MIT	Zero-shot	81.71	107.5	0.76	0.66	No	No	—
89	PatchFMHFCode LITIS	Zero-shot	83.74	77.44	0.81	0.56	No	Yes	—
90	goia-forecast-nano-v0HFCode Gredio	Zero-shot	83.77	76.45	0.82	0.55	No	Yes	—
91	YingLong_300mHFCode Alibaba	Zero-shot	84.64	77.34	0.80	0.55	No	Yes	View model
92	YingLong_110mHFCode Alibaba	Zero-shot	85.33	78.46	0.81	0.56	No	Yes	View model
93	TabPFN-TSHFCode PriorLabs	Zero-shot	85.42	78.15	0.77	0.54	No	Yes	—
94	TempoPFNHFCode University of Freiburg	Zero-shot	85.72	74.90	0.79	0.53	No	Yes	—
95	TinyCastHFCode RAWS Labs	Zero-shot	86.52	82.36	0.77	0.55	No	Yes	—
96	Metamorph1.0-4.5M SRI International	Pretrained	87.38	83.13	0.78	0.55	No	No	—
97	Lingjiang Alibaba Cloud	Zero-shot	88.91	84.79	1.00	0.62	No	No	—
98	YingLong_50mHFCode Alibaba	Zero-shot	89.06	82.24	0.82	0.57	No	Yes	View model
99	Moirai_largeHFCode Salesforce AI Research	Zero-shot	90.19	82.34	0.87	0.60	No	Yes	View model
100	Moirai_baseHFCode Salesforce AI Research	Zero-shot	91.29	82.53	0.90	0.61	No	Yes	View model
101	Metamorph1.0 SRI International	Pretrained	91.67	80.36	0.82	0.55	No	No	—
102	LiteSpecFormerHFCode FlowVortex (Xidian University)	Zero-shot	91.82	88.14	0.87	0.62	No	Yes	—
103	Chronos_largeHFCode AWS AI Labs	Pretrained	94.04	97.24	0.87	0.65	Yes	Yes	View model
104	TimeTron-v2-33MHF Corteri Intelligence	Pretrained	94.91	90.87	0.88	0.61	Yes	No	—
105	Chronos_baseHFCode AWS AI Labs	Pretrained	94.97	98.13	0.88	0.65	Yes	Yes	—
106	PatchTST Princeton University	Deep learning	97.82	89.76	0.85	0.59	No	No	View model
107	FLAIRHFCode Mellon Inc.	Statistical	98.04	87.44	0.84	0.59	No	Yes	—
108	Chronos_smallHFCode AWS AI Labs	Pretrained	98.34	99.09	0.89	0.66	Yes	Yes	—
109	i_transformer Tsinghua University	Deep learning	99.93	92.49	0.89	0.62	No	Yes	—
110	YingLong_6mHFCode Alibaba	Zero-shot	101.1	93.56	0.88	0.61	No	Yes	View model
111	Super-LinearHFCode Ben-Gurion University of the Negev	Pretrained	101.2	114.5	0.86	0.74	Yes	Yes	—
112	TimesFMHFCode Google Research	Pretrained	103.0	97.40	1.08	0.68	Yes	Yes	View family
113	TFT Google Research	Deep learning	103.3	90.54	0.92	0.60	No	No	—
114	VisionTSHF Zhejiang University	Zero-shot	105.5	115.3	0.86	0.75	No	No	—
115	Moirai_smallHFCode Salesforce AI Research	Zero-shot	105.8	95.59	0.95	0.65	No	Yes	View model
116	N-BEATS ServiceNow	Deep learning	106.3	115.8	0.94	0.82	No	No	—
117	FFMHFCode IBM Research	Deep learning	106.9	99.95	1.02	0.70	No	No	—
118	Auto_Arima	Statistical	108.2	114.1	1.07	0.91	No	No	—
119	TTM-R2-PretrainedHFCode IBM Research	Pretrained	109.9	119.5	1.02	0.87	Yes	Yes	View model
120	Auto_ETS	Statistical	111.5	117.3	1.21	7.49	No	No	—
121	Auto_Theta	Statistical	111.8	114.4	1.09	1.24	No	No	—
122	TTM-R1-PretrainedHFCode IBM Research	Pretrained	112.0	119.6	1.08	0.89	Yes	Yes	View model
123	Crossformer Shanghai Jiao Tong University	Deep learning	112.0	101.7	2.57	1.64	No	No	—
124	Seasonal_NaiveHFCode	Statistical	112.8	118.9	1.00	1.00	No	Yes	—
125	DLinear The Chinese University of Hong Kong	Deep learning	114.1	119.2	1.06	0.85	No	No	—
126	DeepAR Amazon Research	Deep learning	115.1	107.7	1.34	0.85	No	No	—
127	TIDE Google Research	Deep learning	115.3	110.0	1.09	0.77	No	No	—
128	Lag-LlamaHF Morgan Stanley & Service Now	Pretrained	119.2	116.3	1.23	0.88	Yes	No	View model
129	NaiveHFCode	Statistical	119.2	124.2	1.27	1.59	No	Yes	—
130	VISIT-1.0	Pretrained	119.4	124.4	1.27	1.59	No	No	—

130 models · Aggregated live from the official GIFT-Eval leaderboard. Refreshes every 12 hours.Last refreshed Sep 12, 2026, 9:20 AM

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: 35(27%)
Zero-shot: 42(32%)
Fine-tuned: 6(5%)
Agentic: 31(24%)
Deep learning: 10(8%)
Statistical: 6(5%)

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 →