> **For coding agents and LLMs:** This is one Social Fetch marketing page (markdown export). For API orientation start with [`/llms.txt`](https://www.socialfetch.dev/llms.txt); for auth and crawl rules use [`/agents.txt`](https://www.socialfetch.dev/agents.txt); for every operation use [`/llms-endpoints.txt`](https://www.socialfetch.dev/llms-endpoints.txt).

## This page

- **On-site (HTML):** <https://www.socialfetch.dev/platforms/hugging-face/datasets>
- **Markdown (.mdx) URL:** <https://www.socialfetch.dev/platforms/hugging-face/datasets.mdx>

---

# Hugging Face Datasets API

Find datasets by keyword, author, language, licence or tags, with download counts and card metadata on every row.

`GET /v1/hugging-face/datasets` returns a page of datasets in `results`. Filter with `query`, `author`, `language`, `license`, `tags` and `gated`, then sort by `downloads`, `likes`, `likes7d`, `likes30d`, `trending`, `created` or `updated`.

Each row carries `id`, `url`, `author`, `likes`, `downloads`, `downloadsAllTime`, `trendingScore`, `gated`, `tags`, `license` and the card metadata the list exposes. `sort=likes30d` needs `query` or `author`, because the platform cannot sort the whole dataset list that way. Page with `page.nextCursor` while `page.hasMore` is true.

## Request

- **Endpoint:** `GET /v1/hugging-face/datasets`
- **Credits:** 1 credit
Failed requests are free
- **Input:** by keyword and filters
- **Full reference:** [docs](https://www.socialfetch.dev/docs/api/v1/hugging-face/datasets/get)
- **Platform page:** [/platforms/hugging-face](https://www.socialfetch.dev/platforms/hugging-face)

## FAQ

### Why is sort=likes30d refused without a query or author?

The platform times out when it sorts the entire dataset list by 30-day likes. The request is refused with a 400 instead of waiting for that timeout.

### Can I filter datasets by language?

Yes, language takes a code such as en. tags takes a comma-separated list that must all match, for example task_categories:text-classification.

### How many results fit on a page?

limit accepts 1 to 1000 and defaults to 50. page.nextCursor carries the rest.

### Does a search return datasets with no public files?

Yes. The list is about the dataset record, not whether its rows can be previewed. Use the rows endpoint to check that.

### How much does a dataset search cost?

Pricing is documented on the operation in the API registry (shown on this page). Confirm on every response with meta.creditsCharged: that field is the billing source of truth. Each paginated page is a separate billed request.