HuggingFace Models Scraper

Pull structured model metadata from the HuggingFace Hub at scale — all-time downloads, likes, trending scores, task, framework, license, gated-access status, and the full repository file manifest for any model across the Hub's 1M+ open-source repositories. Built for ML engineers, AI researchers, and data teams who need to track the open-model ecosystem in a clean spreadsheet without clicking through the Hub one model card at a time.

Why This Scraper?

• 26 machine-learning task filters — narrow to Text Generation, Speech Recognition (ASR), Text-to-Image, Feature Extraction (Embeddings), Object Detection, and 21 more, each from a plain-English dropdown instead of raw API strings.
• 17 framework & format filters — pinpoint models by Transformers, Diffusers, GGUF, Safetensors, ONNX, PEFT (LoRA / adapters), Sentence Transformers, MLX (Apple), and 9 others.
• All-time downloads, likes, and Hub trending score on every row — the three adoption signals you need to rank models by real-world traction, not guesswork.
• License extracted for every model — apache-2.0, mit, llama3.1, cc-by-nc-4.0, and more, pulled from model card data so you can filter by what you're legally allowed to ship.
• Gated-access status detection — each model is flagged open, auto, or manual so you know upfront which weights need an access request before download.
• Full repository file manifest on demand — flip one switch to list every weight, config, tokenizer, and README file per repo, ideal for auditing which models publish GGUF or Safetensors artifacts.
• AND-semantics multi-tag filtering — stack arbitrary Hub tags (gguf, merge, conversational, code) and get back only models that carry every one.
• 5 sort modes — order by trending score, all-time downloads, likes, last-modified, or creation date, with oldest-first paging supported for creation date.
• Up to 50,000 models per run — sweep an entire task, author, or framework in a single pass across the Hub's 1M+ models.

Use Cases

Model Selection & ML Ops

• Shortlist the top text-generation models by downloads before a fine-tuning project
• Filter to a framework your stack supports (Transformers, ONNX, GGUF) to avoid conversion work
• Screen out gated models when you need weights available for automated pipelines
• Compare license terms across candidate models to confirm commercial use is allowed

Market & Competitive Research

• Track every model an organization (meta-llama, google, mistralai) has published
• Measure adoption of a model family by all-time download and like counts
• Benchmark your own released models against competitors in the same task category
• Map which frameworks dominate a given ML task across the Hub

Trend Monitoring

• Pull the trending leaderboard for any task to catch rising models early
• Sort by creation date to surface the newest releases in embeddings or speech
• Watch download velocity on a fixed model set over repeated runs
• Spot new GGUF or MLX quantizations as they appear via tag filters

Dataset & Leaderboard Building

• Assemble a structured catalog of models for internal discovery tools
• Feed a model registry or evaluation harness with fresh metadata
• Build ranked leaderboards by downloads, likes, or trending score
• Power a searchable index of open models for your team

Compliance & License Auditing

• Audit which models in a shortlist carry non-commercial or share-alike licenses
• Flag gated models that require an access agreement before use
• Inventory repository file types to confirm required artifacts (weights, configs) are published
• Document provenance — author, creation date, and last-modified timestamp per model

Getting Started

Search by Keyword

The simplest run — just a search term:

{
    "search": "llama",
    "maxResults": 100
}

Filter by Task and Framework

Combine a task and a framework to get exactly the models your stack can run:

{
    "pipelineTag": "text-generation",
    "libraryName": "gguf",
    "sortBy": "downloads",
    "maxResults": 500
}

One Organization's Full Catalog

Pull every model an organization has published, newest first:

{
    "author": "mistralai",
    "sortBy": "createdAt",
    "sortDirection": "descending",
    "maxResults": 1000
}

Advanced — Tagged, Sorted, with File Lists

{
    "search": "embedding",
    "pipelineTag": "sentence-similarity",
    "libraryName": "sentence-transformers",
    "tags": ["safetensors"],
    "sortBy": "likes",
    "sortDirection": "descending",
    "maxResults": 2000,
    "includeFiles": true
}

Input Reference

Search & Filters

Parameter	Type	Default	Description
search	string	"llama"	Free-text search across model names and descriptions. Leave empty to browse all models by the filters and sort order below.
author	string	null	Only include models from this user or organization (e.g. meta-llama, google, mistralai).
pipelineTag	select	null	Only include models built for one machine-learning task. 26 options including Text Generation, Text Classification, Speech Recognition (ASR), Text-to-Image, Feature Extraction (Embeddings), and Object Detection.
libraryName	select	null	Only include models built with one framework or file format. 17 options including Transformers, Diffusers, GGUF, Safetensors, ONNX, PEFT (LoRA / adapters), and MLX (Apple).
tags	string[]	[]	Extra Hub tags a model must have (e.g. gguf, merge, conversational, code). A model is included only if it has every tag you list.

Results

Parameter	Type	Default	Description
sortBy	select	Trending	Order results by Trending, Most downloads, Most likes, Recently updated, or Recently created.
sortDirection	select	Descending (highest / newest first)	Order direction. The Hub only supports ascending order for "Recently created"; every other sort always returns highest/newest first.
maxResults	integer	100	Maximum number of models to return. Set to 0 to fetch all matches, bounded by a 50,000-model safety cap.
includeFiles	boolean	false	Add each model's repository file list (weights, configs, tokenizer files, README) to the output. Turn off for a lighter, faster result.

Output

Each model is one flat row. Here's a representative result with includeFiles enabled:

{
    "id": "meta-llama/Llama-3.1-8B-Instruct",
    "author": "meta-llama",
    "modelName": "Llama-3.1-8B-Instruct",
    "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct",
    "pipelineTag": "text-generation",
    "libraryName": "transformers",
    "license": "llama3.1",
    "downloads": 4821337,
    "likes": 3894,
    "trendingScore": 42,
    "tags": ["transformers", "safetensors", "llama", "conversational", "license:llama3.1"],
    "gated": "manual",
    "private": false,
    "createdAt": "2024-07-18T16:57:41.000Z",
    "lastModified": "2024-09-25T17:00:57.000Z",
    "sha": "0e9e39f249a16976918f6564b8830bc894c89659",
    "files": ["config.json", "model-00001-of-00004.safetensors", "tokenizer.json", "README.md"],
    "scrapedAt": "2026-07-03T14:30:00Z"
}

Core Fields

Field	Type	Description
id	string	Full model id in author/name form (e.g. meta-llama/Llama-3.1-8B-Instruct).
author	string | null	Owning user or organization.
modelName	string | null	Name portion after the last /.
url	string | null	Canonical model page URL on the Hub.
private	boolean	Whether the repository is private.
scrapedAt	string	ISO timestamp when the row was fetched.

Classification

Field	Type	Description
pipelineTag	string | null	Primary machine-learning task (e.g. text-generation).
libraryName	string | null	Framework or format the model is built with (e.g. transformers).
license	string | null	License id from card data (e.g. apache-2.0, mit, llama3.1).
tags	string[]	All Hub tags — task, language, dataset:, arxiv:, license:, region:, and more.
gated	boolean | string	Gated-access status: false (open), "auto", or "manual".

Metrics

Field	Type	Description
downloads	integer | null	All-time download count.
likes	integer | null	Like count.
trendingScore	number | null	Hub trending score.

Dates & Files

Field	Type	Description
createdAt	string | null	ISO creation timestamp.
lastModified	string | null	ISO last-modified timestamp.
sha	string | null	Latest commit SHA.
files	string[] | null	Repository file names — populated only when includeFiles is on, otherwise null.

Tips for Best Results

• Start with a small maxResults — run 50–100 first to confirm the filters match what you want, then scale up to thousands.
• Combine pipelineTag and libraryName — pairing a task with a framework (e.g. Text Generation + GGUF) is the fastest way to a short, highly relevant list instead of sifting a broad search.
• Page the oldest models first — set sortBy to Recently created with sortDirection ascending to walk a task's history from its very first release forward; this ascending order is only supported on creation date.
• Turn on includeFiles for artifact audits — the file list reveals which repos ship GGUF, Safetensors, or ONNX weights, so you can filter deployable models without opening each card.
• Leave search empty to browse by filters alone — an author, task, or tag combination with no search term returns the full matching set in your chosen sort order.
• Stack tags for precision — every tag you add tightens the result set (all must match), so use it to isolate niche subsets like merge or conversational.
• Sort by downloads for battle-tested picks, trending for what's rising — downloads reward proven adoption, while trending surfaces momentum before a model is widely known.

Pricing

From $1.00 per 1,000 results — a flat rate that makes sweeping an entire task category or organization's catalog affordable. Bronze, Silver, and Gold subscribers pay progressively less; the table below shows total cost at each discount tier.

Results	No discount	Bronze	Silver	Gold
100	$0.12	$0.115	$0.105	$0.10
1,000	$1.20	$1.15	$1.05	$1.00
10,000	$12.00	$11.50	$10.50	$10.00
100,000	$120.00	$115.00	$105.00	$100.00

A "result" is any model row in the output dataset. No compute or time-based charges — you pay per result, plus a small fixed per-run start fee.

Integrations

Export data in JSON, CSV, Excel, XML, or RSS. Connect to 1,500+ apps via:

• Zapier / Make / n8n — Workflow automation
• Google Sheets — Direct spreadsheet export
• Slack / Email — Notifications on new results
• Webhooks — Trigger custom APIs on run completion
• Apify API — Full programmatic access

Legal & Ethical Use

This actor collects publicly available model metadata from the HuggingFace Hub for legitimate research, model discovery, and market analysis. Users are responsible for complying with applicable laws and the HuggingFace Terms of Service. Respect each model's license terms and gated-access requirements before downloading or deploying any weights. Do not use extracted data for any unlawful purpose.