TimesFM 3.0
Release in the TimesFM family · version 3.0 (google/timesfm-3.0-pytorch)
Maintained by Google Research14
TimesFM 3.0 is a Google Research time-series forecasting model released in August 2026. Google's announcement gives it 330 million parameters and describes native multivariate forecasting, support for past-only and past-and-future covariates, and probabilistic forecasts with nine quantiles. The model card lists a 20-layer transformer architecture.174
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- License: TimesFM Non-Commercial License v1.0 (external site: huggingface.co)
- Repository: GitHub repository (external site: github.com)
- Release notes: Google Research blog post (external site: research.google)
Availability and license
Overall availability
Weights are downloadable from Hugging Face without an access gate, but only for non-commercial, non-production use under the TimesFM Non-Commercial License v1.0. Commercial and production use is offered through Google Cloud services such as BigQuery ML.234
Availability is separate from permission: read the license before using or redistributing.
TimesFM Non-Commercial License v1.0 (external site: huggingface.co)13
Apache License 2.0 (timesfm source code) (external site: raw.githubusercontent.com)54
The TimesFM Non-Commercial License v1.0 is issued by Google LLC. It grants a personal, non-transferable, revocable license to use and create derivatives only for non-commercial purposes, which it defines to exclude revenue-generating activity, production systems, interactions with end users, and training or distilling other models for commercial use. It prohibits distributing the model or derivatives. Model outputs are not treated as derivatives, but the restrictions also bar using outputs for commercial or production purposes. The repository notes that earlier TimesFM weights (up to 2.5) remain under Apache 2.0.34
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
The weights are under a license that is not on the rubric's OSI-approved list. Read its terms before use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | Weights (safetensors) are published on Hugging Face without an access gate, under a non-commercial license.21 |
| Inference codeIs code for running the model published? | Public | The Apache-2.0 timesfm package (PyPI "timesfm") includes TimesFM 3.0 inference code with PyTorch and Apple MLX backends; the README gives usage examples.465 |
| Training codeIs the code used to train the model published? | Unknown | The repository includes a fine-tuning example (Transformers with PEFT/LoRA), added in April 2026 before the 3.0 release; no pre-training code for TimesFM 3.0 was found.4 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | The model card lists GiftEvalPretrain (excluding datasets overlapping with fev-bench), Wikipedia pageviews, Google Trends top queries, and synthetic and augmented data. The synthetic data is not documented as released.1 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Unknown | The model card lists architecture settings, and the blog post describes the approach at a high level; no technical report for version 3.0 was found.17 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The README and blog report benchmark standings on fev-bench, TIME, and GIFT-Eval; code to re-run those evaluations for this release was not identified.47 |
What it is useful for
Run and use notes
- The README documents installation with "pip install timesfm[torch]" or "pip install timesfm[mlx]" (MLX-native inference on Apple silicon without PyTorch) and notes that contexts longer than 15,360 points are truncated to the most recent points.4
Organization context
Provenance and derivatives
Trained by Google Research. The model card lists training data from GiftEvalPretrain, Wikipedia pageviews, Google Trends, and synthetic data. No other base model is named.1
Other releases in the TimesFM family
- TimesFM 2.5 200MModel-disclosure tier (USASI rubric v0.1): Open-weight
In the news
Google releases TimesFM 3.0 weights under a non-commercial license
Google Research released TimesFM 3.0, a time-series forecasting model, in August 2026 under the TimesFM Non-Commercial License v1.0 instead of the Apache 2.0 license used for earlier TimesFM weights. The license allows use of the weights only for non-commercial purposes and prohibits redistributing them; the repository says commercial and production use of TimesFM 3.0 is permitted through authorized Google Cloud services such as BigQuery ML. Weights up to version 2.5 and the TimesFM source code remain under Apache 2.0.
U.S. eligibility
Eligible · basis: Documented U.S. control
The model card states TimesFM was developed by Google Research, and the weights license is issued by Google LLC. Alphabet Inc.'s fiscal 2025 Form 10-K lists its principal executive offices in Mountain View, California, and Exhibit 21.01 lists Google LLC as a Delaware subsidiary of Alphabet.1389
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.