# Lichess Rated Standard Chess Games Database

## Facts

- Provider: Lichess (lichess.org)
- License: CC0-1.0 (https://database.lichess.org/)
- Size: 26,140 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/Lichess/standard-chess-games
- Access: public, readable without an account
- Category: Machine Learning & AI
- Source: https://database.lichess.org/#standard_games
- Page: https://scigantic.com/archive/c54b8669-dc51-5cda-8a86-219c541bdcfe
- Keywords: Machine Learning & AI, Parquet, Hugging Face

## About

Every rated standard chess game played on lichess.org, the free open-source chess server, converted from the monthly PGN database dumps into Parquet for analysis. The card reports 6,771,826,271 games, updated monthly. Data are hive-partitioned by year and month (data/year=2013 through data/year=2025 at the time of listing, measured 2026-11-14, each month split into several Parquet shards); the repository uses about 5.1 TB. Each row is one game with the PGN header fields (players, Glicko-2 ratings in the WhiteElo and BlackElo tags, result, time control, opening, termination) and a movetext column that carries clock times as %clk comments since April 2017; about 6% of games include Stockfish %eval annotations. The data are used to study human decision-making and skill, rating systems, time pressure, opening theory and cheating detection, and to train and evaluate human-like chess models such as Maia and move-prediction transformers. Read partitions with Hugging Face datasets (streaming=True), DuckDB, polars or pyarrow, filtering on year and month; the card shows how to turn a subset back into PGN for python-chess. Lichess releases its database exports under CC0, free for research, commercial use and redistribution.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("Lichess/standard-chess-games")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/year=2013/month=01/train-00000-of-00001.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/c54b8669-dc51-5cda-8a86-219c541bdcfe
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/c54b8669-dc51-5cda-8a86-219c541bdcfe
- Over MCP: `get_schema_card` with id `c54b8669-dc51-5cda-8a86-219c541bdcfe` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Machine Learning & AI datasets](https://scigantic.com/topics/machine-learning-and-ai)
