yusp48/geometry-dash-levels

Geometry Dash Level Dataset

Subsets

2024_300k

Dump of ~300k levels from the Geometry Dash servers, sorted by the number of likes. 66 JSONL shards (~660MB each, ~43GB total).

Files: 2024_300k/levels-v1-00000.jsonl through 2024_300k/levels-v1-00065.jsonl

2026_50k_rated

~50k rated/featured levels scraped from the Geometry Dash servers in February 2026. 39 JSONL shards (~500MB each, ~19GB total).

Files: 2026_50k_rated/levels-v2-00000.jsonl through 2026_50k_rated/levels-v2-00038.jsonl

2024_300k_nodata

Metadata-only version of 2024_300k with the level_string field removed. Contains all other fields (level name, ID, likes, downloads, difficulty, stars, etc.) in a single merged JSONL file. 330,925 levels, ~261MB.

File: 2024_300k_nodata/2024_300k_nodata.jsonl

2026_50k_rated_nodata

Metadata-only version of 2026_50k_rated with the level_string field removed. Contains all other fields in a single merged JSONL file. 53,183 levels, ~28MB.

File: 2026_50k_rated_nodata/2026_50k_rated_nodata.jsonl

About the _nodata subsets

The _nodata subsets are lightweight versions of the full datasets that contain only level metadata (name, difficulty, stars, likes, downloads, game version, etc.) without the actual level geometry data (level_string). This makes them much smaller and faster to download — useful for exploratory analysis, filtering, or building indexes without needing the full ~62GB of level data.

Level String Encoding

In both full subsets, each record's level_string is stored in the same encoding Geometry Dash uses for save files:

urlsafe_base64(gzip(raw_level_string_utf8))

To decode back to the raw GD level string:

import base64, gzip

def decode_level_string(s: str) -> str:
    return gzip.decompress(base64.urlsafe_b64decode(s.encode("ascii"))).decode("utf-8")

Notes

Passwords are not decoded. Refer to the gddocs XOR docs if you need to decode passwords.

Thanks to https://wyliemaster.github.io/gddocs/ for documentation of Geometry Dash formats.