TimeGEN-1
TimeGEN-1
Version: 1
NixtlaLast updated October 2025

Models from Microsoft, Partners, and Community

Models from Microsoft, Partners, and Community models are a select portfolio of curated models both general-purpose and niche models across diverse scenarios by developed by Microsoft teams, partners, and community contributors
  • Managed by Microsoft: Purchase and manage models directly through Azure with a single license, world class support and enterprise grade Azure infrastructure
  • Validated by providers: Each model is validated and maintained by its respective provider, with Azure offering integration and deployment guidance.
  • Innovation and agility: Combines Microsoft research models with rapid, community-driven advancements.
  • Seamless Azure integration: Standard Microsoft Foundry experience, with support managed by the model provider.
  • Flexible deployment: Deployable as Managed Compute or Serverless API, based on provider preference.
Learn more about models from Microsoft, Partners, and Community

Key capabilities

About this model

TimeGEN-1 is a generative pre-trained forecasting and anomaly detection model for time series data. The model excels at zero-shot forecasting by leveraging temporal correlations learnt on billions of time series.

Key model capabilities

  • Zero-shot forecasting for new time series without training
  • Anomaly detection
  • Fine-tuning capabilities on new data to improve accuracy
  • Support for multivariate input
  • Exogenous variables support
  • Demand forecasting

Use cases

See Responsible AI for additional considerations for responsible use.

Key use cases

TimeGEN-1 can produce accurate forecasts for new time series without training using only historical values and exogenous covariates as inputs. The model is optimized for forecasting and anomaly detection tasks, including demand forecasting, monitoring ordered data points to spot irregularities that may signal issues or threats, and leveraging exogenous variables as external factors that can influence forecasts.

Out of scope use cases

  • Transferability between Domains: Time series models trained on data from different domains, may not always perform accurately when applied to a different domain.
  • Impact of Extreme Events: Extreme events (such as natural disasters, economic crises, or pandemics) can significantly impact the accuracy of time series models. These events often create patterns and trends that the model has not encountered during training, leading to poor performance.
  • Data Quality and Preprocessing: The quality of the input data greatly affects the accuracy of the time series model. Issues such as missing values, outliers, and inconsistent data can lead to unreliable predictions.
  • Accuracy: Artificial intelligence and machine learning are rapidly evolving fields of study. We are constantly working to improve TimeGEN-1 to make them more accurate, reliable, safe, and beneficial. Given the probabilistic nature of machine learning, the use of our Product may, in some situations, result in incorrect Output. You should always evaluate the accuracy of any Output as appropriate for your use case, including by using human review of the Output.

Pricing

Pricing is based on a number of factors, including deployment type and tokens used. See pricing details here.

Technical specs

TimeGEN-1 was pretrained on 100 billion tokens of time series data from publicly available sources. The model has 500M parameters.

Training cut-off date

The pretraining data has a cutoff of September 2023 but some tuning data is more recent up to March 2024.

Training time

TimeGEN-1 was trained between July 2023 and October 2023.

Input formats

Time series data as json or dataframes (Support for multivariate input).

Output formats

Time Series data as json.

Supported languages

The provider has not supplied this information.

Sample JSON response

{
  "timestamp": [
    "2016-01-14 00:00:00",
    "2016-01-15 00:00:00",
    "2016-01-16 00:00:00",
    "2016-01-17 00:00:00",
    "2016-01-18 00:00:00",
    "2016-01-19 00:00:00",
    "2016-01-20 00:00:00"
  ],
  "value": [
    8.05051040649414,
    7.909270763397217,
    7.8442702293396,
    8.21353816986084,
    8.600921630859375,
    8.293558120727539,
    8.150216102600098
  ],
  "input_tokens": 43,
  "output_tokens": 7,
  "finetune_tokens": 0
}

Model architecture

TimeGEN-1 is an auto-regressive time series model optimized for forecasting and anomaly detection tasks. The model excels at zero-shot forecasting by leveraging temporal correlations learnt on billions of time series. TimeGEN-1's parameters can be fine-tuned on new data to further improve accuracy.

Long context

The provider has not supplied this information.

Optimizing model performance

TimeGEN-1's parameters can be fine-tuned on new data to further improve accuracy.

Additional assets

The provider has not supplied this information.

Training disclosure

Training, testing and validation

TimeGEN-1 was pretrained on 100 billion tokens of time series data from publicly available sources. We used custom training libraries Nixtla's open-source libraries, Nixtla's Research Cluster and production clusters for pretraining. Fine-tuning and evaluation were also performed on third-party cloud compute.

Distribution

Distribution channels

The provider has not supplied this information.

More information

Responsible AI considerations

Safety techniques

The provider has not supplied this information.

Safety evaluations

The provider has not supplied this information.

Known limitations

  • Accuracy: Artificial intelligence and machine learning are rapidly evolving fields of study. We are constantly working to improve TimeGEN-1 to make them more accurate, reliable, safe, and beneficial. Given the probabilistic nature of machine learning, the use of our Product may, in some situations, result in incorrect Output. You should always evaluate the accuracy of any Output as appropriate for your use case, including by using human review of the Output.
  • Transferability between Domains: Time series models trained on data from different domains, may not always perform accurately when applied to a different domain.
  • Impact of Extreme Events: Extreme events (such as natural disasters, economic crises, or pandemics) can significantly impact the accuracy of time series models. These events often create patterns and trends that the model has not encountered during training, leading to poor performance.
  • Data Quality and Preprocessing: The quality of the input data greatly affects the accuracy of the time series model. Issues such as missing values, outliers, and inconsistent data can lead to unreliable predictions.
Mitigations: Ensure thorough data preprocessing, including cleaning, normalization, and imputation of missing values. Regularly audit the data pipeline to maintain high data quality.

Acceptable use

Acceptable use policy

The provider has not supplied this information.

Quality and performance evaluations

Source: Nixtla The provider has not supplied this information.

Benchmarking methodology

Source: Nixtla The provider has not supplied this information.

Public data summary

Source: Nixtla The provider has not supplied this information.
Model Specifications
LicenseCustom
Training DataOctober 2023
Last UpdatedOctober 2025
ProviderNixtla
Languages1 Language