Time-Series Foundation Models

Pretrained forecasters used zero-shot or fine-tuned.

Tier Grades

Tier A: Frontier AI labs

feasible

medium confidence

Amazon is the developer.

Tier B: Indonesian AI labs

feasible

medium confidence

Models of 20M to 710M parameters are small, though training compute is not given in the notes.

Tier C: Institution units such as ITB

feasible

medium confidence

Models of 20M to 710M parameters are small, though training compute is not given in the notes.

Tier D: Organisations such as SEMAI (PoT, PoC, PoV)

feasible

medium confidence

Zero-shot use of a pretrained forecaster on unseen data fits proof-of-concept work; training compute is not given.

Where this comes from

Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.

Anchors

ComputeChronos: T5-based, 20M to 710M parameters; training compute not given.
DataPublic plus synthetic Gaussian-process data; 42-dataset benchmark.
Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): not sourced in the notes.
Time horizonMonths.
Shelf-life signalsnot in the notes
Pipeline roleTier A and big tech release pretrained forecasters; Tier C benchmarks and fine-tunes them; Tier D applies them zero-shot in proofs of value.

Where this comes from

Team size is a proxy (author counts of exemplar papers), not a headcount.

Who and What

Exemplars

Chronos

Observed actors

Amazon

Linked models

+38

ChatTimetime seriesproduced by ChatTime-1-7B-Basetime seriesproduced by ChatTime-1-7B-Chattime seriesproduced by

MOMENT (MOMENT-1-large)time seriesproduced by

Granite TinyTimeMixer (TTM) r2time seriesproduced by

Lag-Llamatime seriesproduced by

TimeGPT-1time seriesproduced by

Time-LLMtime seriesproduced by

MSTLtime seriesused by

Lingjiang2.0time seriesproduced by

Chronos-Bolttime seriesproduced by

Xihe-ultratime seriesproduced by

Toto 1.0time seriesproduced by

QiYao-Mtime seriesproduced by

STRIDE_w_Synapsetime seriesproduced by

TimesFMtime seriesproduced by

TimesFM 1.0time seriesproduced by

TimesFM 2.0time seriesproduced by

Pangu Prediction Modeltime seriesproduced by

Granite FlowStatetime seriesproduced by

Granite PatchTST-FM r2time seriesproduced by

Granite TTM-R1time seriesproduced by

Granite TinyTimeMixer (TTM) r2.1time seriesproduced by

EXAONE-Forecasttime seriesproduced by

TimeGPTtime seriesproduced by

TimeGPT-2time seriesproduced by

TimeGPT-2.1time seriesproduced by Croston methodtime seriesused by ETS (exponential smoothing)time seriesused by

TabPFN-TStime seriesproduced by

Moirai 1.0time seriesproduced by

Moirai 2.0 basetime seriesproduced by

Moirai 2.0 largetime seriesproduced by

Moirai 2.0 R smalltime seriesproduced by Seasonal_Naivetime seriesused by Theta methodtime seriesused by

DeOSAlphaTimeGPTPredictor-2025time seriesproduced by

ChakraTStime seriesproduced by

Links

R&D TopicTime-Series Foundation Models

Models

ChronosAmazon, time series

Lingjiang2.0Alibaba (Qwen), time series

Chronos-2Amazon, time series

Chronos-BoltAmazon, time series+58 more

Use CasesIndustrial Forecasting (Load, Throughput, Demand)Industrial Demand ForecastingRetail and Logistics Anomaly Detection on Sensor, SCADA and Historian DataIndustrial Energy OptimisationIndustrial+1 more

Who Runs It

PT Amman Mineral Nusa Tenggara (AMMAN)In production, 8 Jul 2026

PT Pertamina Hulu Rokan (PHR)In production, 17 Oct 2025 MIND IDResearch, 14 Oct 2025

Shenzhen Energy GroupIn production, 23 Jun 2025+4 more

Graph

Models 62

Moirai 2.0Salesforce, time seriesproduced by

Pangu Prediction ModelHuawei (Huawei Cloud), time seriesproduced by

Toto 2.0Datadog, time seriesproduced by

TimesFM 2.5Google, time seriesproduced by

Chronos-2Amazon, time seriesproduced by

Granite TinyTimeMixer (TTM) r2IBM, time seriesproduced by

TimesFMGoogle, time seriesproduced by

Granite TinyTimeMixer (TTM) r2.1IBM, time seriesproduced by

Lag-LlamaMorgan Stanley, ServiceNow (ServiceNow Research, Montreal), Universite de Montreal, Mila, time seriesproduced by

TimeGPTNixtla, time seriesproduced by

TimeGPT-2Nixtla, time seriesproduced by

TimesFM 2.0Google, time seriesproduced by

+50

TimesFM 3.0Google, time seriesproduced by ChatTime-1-7B-Basetime seriesproduced by

ChronosAmazon, time seriesexemplar

Chronos-BoltAmazon, time seriesproduced by

Granite PatchTST-FM r1IBM, Rensselaer Polytechnic Institute, time seriesproduced by

Granite PatchTST-FM r2IBM, Rensselaer Polytechnic Institute, time seriesproduced by

Granite TimeSeries TTM r3IBM, time seriesproduced by

Granite TSPulseIBM, time seriesproduced by Kronostime seriesproduced by

MOMENT (MOMENT-1-large)Carnegie Mellon University, time seriesproduced by

TimeGPT-1Nixtla, time seriesproduced by

TimeGPT-2.1Nixtla, time seriesproduced by

TimesFM 1.0Google, time seriesproduced by ChatTimetime seriesproduced by ChatTime-1-7B-Chattime seriesproduced by

Granite FlowStateIBM, time seriesproduced by

Granite TTM-R1IBM, time seriesproduced by

Moirai 1.0Salesforce, time seriesproduced by

Moirai 2.0 baseSalesforce, time seriesproduced by

Moirai 2.0 largeSalesforce, time seriesproduced by

Moirai 2.0 R smallSalesforce, time seriesproduced by

Moirai 2.0 smallSalesforce, time seriesproduced by

MSTLUniversity of Melbourne, time seriesused by

QiYao-MECNU, Huawei (Huawei Cloud), time seriesproduced by

TabPFN-TSPrior Labs, time seriesproduced by

Time-LLMMonash University, Ant Group, time seriesproduced by

Timer-S1Tsinghua University, ByteDance, time seriesproduced by

TiRexNX-AI, time seriesproduced by

TiRex-2NX-AI, time seriesproduced by

Toto 1.0Datadog, time seriesproduced by

Toto 2.0 1BDatadog, time seriesproduced by

Toto 2.0 2.5BDatadog, time seriesproduced by

Toto 2.0 22MDatadog, time seriesproduced by

Toto 2.0 313MDatadog, time seriesproduced by

Toto 2.0 4MDatadog, time seriesproduced by

ChakraTSYHat Labs, time seriesproduced by Croston methodtime seriesused by

DeOSAlphaTimeGPTPredictor-2025vencortex, time seriesproduced by ETS (exponential smoothing)time seriesused by

EXAONE-ForecastLG AI Research, time seriesproduced by

FlowState r1.1IBM, time seriesproduced by Kronos-basetime seriesproduced by Kronos-minitime seriesproduced by Kronos-smalltime seriesproduced by

Lingjiang2.0Alibaba (Qwen), time seriesproduced by Seasonal_Naivetime seriesused by

STRIDE_w_SynapseGoogle, time seriesproduced by

Sundial (Timer 3.0) base-128mTsinghua University, time seriesproduced by Theta methodtime seriesused by

Xihe-ultraAnt Group, time seriesproduced by

Papers

A decoder-only foundation model for time-series forecastingAbhimanyu Das et al., arXiv preprint, 2023 founding
Chronos: Learning the Language of Time SeriesAbdul Fatir Ansari et al., arXiv preprint, 2024 state of the art
Unified Training of Universal Time Series Forecasting TransformersGerald Woo et al., arXiv preprint, 2024 state of the art
Lag-Llama: Towards Foundation Models for Probabilistic Time Series ForecastingKashif Rasul et al., arXiv preprint, 2023 state of the art
Foundation Models for Time Series Analysis: A Tutorial and SurveyYuxuan Liang et al., KDD'24 (per arXiv comment), 2024 survey

Where this comes from

The papers to read first: where the direction began, its current best, and a survey when one exists.

Sources

Chronos: Learning the Language of Time Series arxiv.org https://arxiv.org/abs/2403.07815

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Time-Series Foundation Models", https://aurion.semai.id/topics/time-series-foundation-models/, records of October 2026.