Datasets:

nuriyev
/

lichess-elite

license: cc0-1.0
task_categories:
  - text-generation
  - reinforcement-learning
language:
  - en
tags:
  - chess
  - games
  - lichess
  - pgn
  - openings
pretty_name: Lichess Elite Database
size_categories:
  - 10M<n<100M

Lichess Elite Database

A curated collection of high-level chess games from Lichess.org, extracted from the Lichess Elite Database by nikonoel.

Dataset Description

This dataset contains 26.3 million games played by strong players on Lichess, filtered to include only:

White player: rated 2400+ (2500+ from Dec 2021)
Black player: rated 2200+ (2300+ from Dec 2021)
Time controls: Rapid and Blitz (bullet excluded)
Variant: Standard chess only

Source

Original data compiled by nikonoel from the Lichess Open Database.

Dataset Structure

Column	Type	Description
`event`	string	Game event (e.g., "Rated Blitz game")
`site`	string	Lichess game URL
`white`	string	White player username
`black`	string	Black player username
`result`	string	Game result: `1-0`, `0-1`, `1/2-1/2`
`utcdate`	string	Game date (YYYY.MM.DD)
`utctime`	string	Game start time (HH:MM:SS)
`whiteelo`	int	White player's rating
`blackelo`	int	Black player's rating
`whiteratingdiff`	string	White's rating change
`blackratingdiff`	string	Black's rating change
`eco`	string	ECO opening code
`opening`	string	Opening name
`timecontrol`	string	Time control (e.g., "180+0", "600+0")
`termination`	string	How game ended (Normal, Time forfeit, etc.)
`moves`	string	Full game in SAN notation

Use Cases

Opening preparation: Study how strong players handle specific openings
Chess engine training: Train or fine-tune chess models
Game analysis: Statistical analysis of high-level play
Move prediction: Sequence modeling for next-move prediction
Style analysis: Study playing patterns of titled players

Usage

from datasets import load_dataset

dataset = load_dataset("nuriyev/lichess-elite")

# Filter by opening
sicilian = dataset["train"].filter(lambda x: x["eco"].startswith("B"))

# Filter by rating
super_gm = dataset["train"].filter(lambda x: x["whiteelo"] >= 2700)

Citation

If you use this dataset, please credit the original source:

@misc{lichess_elite_database,
  author = {nikonoel},
  title = {Lichess Elite Database},
  year = {2020-2025},
  url = {https://database.nikonoel.fr/}
}

License

The original Lichess game data is released under CC0 1.0.

Acknowledgments

nikonoel for curating and maintaining the Lichess Elite Database
Lichess.org for making game data freely available