CHRONOS is a simple yet effective framework for pretrained probabilistic time series models. It tokenizes time series values using scaling and quantization into a fixed vocabulary and trains existing transformer-based language model architectures on these tokenized time series via the cross-entropy loss. CHRONOS models are pretrained on a large collection of publicly available datasets, complemented by a synthetic dataset generated via Gaussian processes to improve generalization. In a comprehensive benchmark of 42 datasets, CHRONOS models significantly outperform other methods on datasets that were part of the training corpus and have comparable or superior zero-shot performance on new datasets relative to methods trained specifically on them. CHRONOS models leverage time series data from diverse domains to improve zero-shot accuracy on unseen forecasting tasks, positioning pretrained models as a viable tool to simplify forecasting pipelines. CHRONOS tokenizes time series into discrete bins through simple scaling and quantization of real values, allowing the training of off-the-shelf language models on this "language of time series" without changes to the model architecture. This approach proves effective and efficient, highlighting the potential for language model architectures to address a broad range of time series problems with minimal modifications. The scarcity of publicly available time series datasets is more critical than the modeling framework for developing a useful general-purpose time series forecasting model. CHRONOS integrates data augmentation strategies, including TSMixup and KernelSynth, to address the limitations of small training datasets, enhancing model robustness and generalization. CHRONOS achieves impressive zero-shot forecasting performance out of the box without task-specific adjustments. Its accuracy and relatively modest model size position it as a preferable alternative to larger, more computationally demanding models for zero-shot forecasting applications. CHRONOS can seamlessly integrate with future advancements in large language models (LLMs), making it an ideal candidate for further development as a generalist time series model. The paper is organized into sections discussing background, related work, the CHRONOS framework, data augmentation techniques, experimental results, and future directions. The results show that CHRONOS models perform well on both in-domain and zero-shot forecasting tasks, outperforming traditional models and task-specific deep learning approaches. The models are trained on a large collection of time series data, tokenized via scaling and quantization, and show significant improvements in performance across various datasets. The results highlight the effectiveness of CHRONOS as a generalist time series forecasting model.CHRONOS is a simple yet effective framework for pretrained probabilistic time series models. It tokenizes time series values using scaling and quantization into a fixed vocabulary and trains existing transformer-based language model architectures on these tokenized time series via the cross-entropy loss. CHRONOS models are pretrained on a large collection of publicly available datasets, complemented by a synthetic dataset generated via Gaussian processes to improve generalization. In a comprehensive benchmark of 42 datasets, CHRONOS models significantly outperform other methods on datasets that were part of the training corpus and have comparable or superior zero-shot performance on new datasets relative to methods trained specifically on them. CHRONOS models leverage time series data from diverse domains to improve zero-shot accuracy on unseen forecasting tasks, positioning pretrained models as a viable tool to simplify forecasting pipelines. CHRONOS tokenizes time series into discrete bins through simple scaling and quantization of real values, allowing the training of off-the-shelf language models on this "language of time series" without changes to the model architecture. This approach proves effective and efficient, highlighting the potential for language model architectures to address a broad range of time series problems with minimal modifications. The scarcity of publicly available time series datasets is more critical than the modeling framework for developing a useful general-purpose time series forecasting model. CHRONOS integrates data augmentation strategies, including TSMixup and KernelSynth, to address the limitations of small training datasets, enhancing model robustness and generalization. CHRONOS achieves impressive zero-shot forecasting performance out of the box without task-specific adjustments. Its accuracy and relatively modest model size position it as a preferable alternative to larger, more computationally demanding models for zero-shot forecasting applications. CHRONOS can seamlessly integrate with future advancements in large language models (LLMs), making it an ideal candidate for further development as a generalist time series model. The paper is organized into sections discussing background, related work, the CHRONOS framework, data augmentation techniques, experimental results, and future directions. The results show that CHRONOS models perform well on both in-domain and zero-shot forecasting tasks, outperforming traditional models and task-specific deep learning approaches. The models are trained on a large collection of time series data, tokenized via scaling and quantization, and show significant improvements in performance across various datasets. The results highlight the effectiveness of CHRONOS as a generalist time series forecasting model.