Back to Gap Map Infrastructure / Model hubs & registries

Kaggle Models

closed / Overall score: n/a

Model hub inside Kaggle, Google's data science platform, where users and organizations publish models as variations per framework and size, each variation versioned as file snapshots. Models are private until published, searchable and sortable by downloads or votes, and downloadable by handle through the kagglehub library and the Kaggle CLI. TensorFlow Hub moved here in 2023, and tfhub.dev links redirect to it.

kaggle.com served a browser challenge to every fetch, so this entry is read from Kaggle's own CLI and client documentation. TensorFlow Hub was never a separate entry on the map.

Openness

1 high confidence

1.0

license: proprietary(hosted service
source: closed(hosted service

Kaggle Models is part of Kaggle's hosted platform, and what Kaggle publishes openly are two clients for it, a Python library and a CLI. Neither contains the service, so nothing about the platform can be run elsewhere.

https://raw.githubusercontent.com/Kaggle/kaggle-api/HEAD/README.md recorded 2026-09-27
"The official CLI to interact with Kaggle." Another client; neither repository contains the platform.
https://raw.githubusercontent.com/Kaggle/kagglehub/HEAD/README.md recorded 2026-09-27
"The kagglehub library provides a simple way to interact with Kaggle resources such as datasets, models, notebook outputs in Python." A client of the hosted service.

Adoption

not assessed

Kaggle's model listing API and pages served a browser challenge on every attempt, and the CLI documentation describes paged results with no total. No usage figure could be read, and the download counts of the kagglehub client would measure a client rather than the hub.

https://raw.githubusercontent.com/Kaggle/kaggle-api/HEAD/docs/models.md recorded 2026-09-27
The models list command pages with "--page-token" and sorts by "downloadCount" among others; no total count is documented.

Capability

4 medium confidence

4.0

Any user or organization can publish a model, split it into variations by framework, and add versioned snapshots that kagglehub downloads by handle. The hub holds models only, with datasets and notebooks living elsewhere on Kaggle and no Git history, so it sits a step below Hugging Face Hub.

https://raw.githubusercontent.com/Kaggle/kaggle-api/HEAD/docs/model_variations_versions.md recorded 2026-09-27
"Each version represents a snapshot of the model variation files at a point in time."
https://raw.githubusercontent.com/Kaggle/kaggle-api/HEAD/docs/models_metadata.md recorded 2026-09-27
"1. The model 2. The variations 3. The variation versions"; "ownerSlug: the slug of the user or organization"; frameworks "tensorFlow1, tensorFlow2, tfLite, tfJs, pyTorch, jax, coral, ...".
https://raw.githubusercontent.com/Kaggle/kagglehub/HEAD/README.md recorded 2026-09-27
"kagglehub.model_download('google/bert/tensorFlow2/answer-equivalence-bem/1')" and "kagglehub.model_upload(handle, local_model_dir, version_notes='improved accuracy')".
https://raw.githubusercontent.com/tensorflow/hub/HEAD/README.md recorded 2026-09-27
"TensorFlow Hub has moved to Kaggle Models" and "Starting November 15th 2023, links to tfhub.dev redirect to their counterparts on Kaggle Models."

Verified 2026-09-27