Job market datasets: 7 real sources for job postings data (2026)

A job market dataset request usually means one of two things: a static file of job postings to analyze (for research, an ML model, or a class project), or a continuously updated feed to build on. The sources below cover both, from free Kaggle snapshots to commercial bulk vendors - with the honest catch attached to each.

1. Kaggle snapshots

Kaggle hosts several widely used job-postings datasets - the LinkedIn job postings dataset (roughly 124,000 postings with salary, company, and skill fields) is the most cited, alongside assorted Indeed and Glassdoor scrapes. Free, instantly downloadable, and fine for coursework and model prototyping. The catch: every Kaggle dataset is a frozen snapshot of one moment, usually one source, collected with undocumented methodology. Anything time-sensitive built on it is already wrong.

2. The USAJOBS archive

US federal postings are the cleanest fully free longitudinal option: the official USAJOBS API serves current openings, and community projects have accumulated historical archives by polling it over time. Structured salary on every record, but federal-government scope only.

3. Adzuna aggregates

Adzuna’s free API tier includes derived-data endpoints - salary histograms, job counts by region and category, salary history over time - that function as a ready-made macro dataset for 16 countries. Good for trend charts; there is no bulk export of the underlying postings. Our Adzuna comparison covers the per-job thinness.

4. Official statistics

BLS JOLTS, OEWS, and FRED serve rigorous survey-based series free - the right base layer for research that needs methodology a reviewer will accept. They are aggregates, not postings: no company names, no individual records. Our labor market data sources post maps this landscape in full.

5. Commercial bulk vendors

Coresignal (postings back to 2020, delivered as Parquet, CSV, or JSONL), Lightcast, LinkUp, and Revelio Labs sell large historical postings datasets with enterprise procurement attached - typically four-to-six-figure engagements. This is the right tier for training corpora and longitudinal research at scale. Our Coresignal comparison covers when that trade makes sense.

6. Single-vertical samples

For some verticals, a curated sample beats a giant general file. Our own Upwork jobs dataset is one example - a structured sample of freelance-market postings with the first rows free.

7. Build the dataset from an API

The option most analyses actually need: define the slice (roles, locations, sources, date range), pull it from a jobs API, and export exactly that - reproducibly, with a refresh path when the data ages.

curl https://api.jobspipe.dev/v1/jobs/search \
  -H "Authorization: Bearer jp_live_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{ "job_title_or": ["machine learning"], "posted_at_gte": "2026-06-13", "limit": 100 }' \
  > ml_jobs_page1.json

JobsPipe records come normalized across 30+ ATS sources - structured salary, resolved locations, cross-source dedup, explicit source attribution - which is most of the cleaning work a scraped dataset forces on you. Twelve months of expired-posting retention covers recent-history questions.

Static file or living feed?

The decision rule is the question you are answering. Historical and one-off questions (“how did postings shift in 2024?”) fit static datasets - Kaggle free, commercial vendors at depth. Current and recurring questions (“who is hiring for this now?”) need a feed, because a dataset of live postings starts rotting the day it is exported. Most teams that start with a file end up wanting the API underneath it. If you are buying live data rather than a static file, our where to get job posting data comparison covers that market.