Kaggle: competitions, public datasets and 30 free GPU hours every week

Kaggle is a Google-owned data science and machine learning platform: it hosts prize-money competitions, thousands of public datasets, browser-based notebooks and short courses with free certificates. Nothing costs a cent, including GPUs capped at 30 hours a week. The community has grown past 30 million members, from students to frontier research labs. Everything runs in English, which rarely slows anyone down since the real language here is Python.

Competitions, notebooks and datasets: how Kaggle actually works

Kaggle rests on three pillars: machine learning competitions posted by companies, public datasets to explore, and Jupyter notebooks that run online in Python or R. Founded in 2010 by Anthony Goldbloom, the platform joined Google in 2017 and has served as the training ground of the global data community ever since.

A competition fits in one sentence: an organization publishes a real problem with its data, you submit predictions, and a public leaderboard ranks everyone. Some contests hand out tens of thousands of dollars. Others only award medals, and those medals often carry more weight on a resume than the cash.

Community Benchmarks and hackathons: Kaggle now grades the models too

Kaggle hosts AI model evaluation benchmarks alongside its classic competitions, and the site now bills itself as the world's AI proving ground. With Community Benchmarks, anyone can design custom tests, run them against models from Google, Anthropic or DeepSeek within free quotas, then publish a leaderboard.

Hackathons follow the same open logic. A school or a company can set one up for free, with Kaggle funding up to $10,000 in prizes. Google DeepMind even ran a $200,000 benchmark-design contest on AGI there, a sign that major labs take the platform seriously.

• Build your own benchmark and share it with the community
• Test frontier models for free, within quota limits
• Host a hackathon for your school or company

Free GPUs, short courses: getting up and running

A free account unlocks everything. You open a notebook, switch on a T4 or P100 GPU in the side panel, and training starts with zero setup (there can be a queue at peak times, so heavy jobs run best in the morning). The in-house courses, from Python to computer vision, take a few hours each and come with a certificate, a handy add-on to longer AI learning resources.

Compute quotas are predictable, a welcome change from free services that cut you off without warning. They do shift over time, though: Kaggle's official documentation is the reference for current numbers.

Frequently asked questions