happy programmer's day. compress challenge.b64 as small as you can.
this is a Hutter Prize style contest, but not for general compression.
- ai is allowed, but dumping the file into a model is expensive. finding the structure is what pays.
- any language. You ship a repo with a
Dockerfilethat builds and runs your answer.
- it's theoretically possible to compress this file from 64mb to about 2mb (32x)
- the challenge ends on 19/09/2026
- the prize is awarded to the best algorithm and is calculated like this:
prize = 256,000 * log2(size(original file) / size(compressed file))
so if you compress the file 32x you will be awarded 1,280,000 tomans
score = size(compressed output) + size(your repo except Dockerfile and .git)
lower is better. Libraries, payloads, extra files all count.
docker run --network none IMAGE --compress /data/in /data/out
docker run --network none IMAGE --decompress /data/out /data/back
/data/back must match the original file byte for byte.
build and run have no network. nothing is downloaded.
FROM only works if that image is already on the machine.
the public runner has python:3.12-alpine, alpine:3.21, gcc:14-bookworm, golang:1.23-alpine, and rust:1.83-alpine. Anything else: FROM scratch and ship the files (they count).
--compress and --decompress have 60 seconds each. The build has 5 minutes. Clone has 2 minutes. Over that, a mismatch, or a network call at build/run is a fail.
- put a
Dockerfilein your own public repo (root, or setdir). - fork this repo, add yourself to participants.json, open a PR.
{
"name": "your-github-username",
"repository": "https://github.com/you/your-256-repo"
}optional: "branch": "main", "dir": "subdir".
256-example-solution embeds the file and writes an empty compressed output. valid, terrible score.
hint 1
besides other patterns, some parts may be mathematical sequences. ask your agent to help find them.
# your solution repo:
# Dockerfile + program
# entrypoint: --compress /data/in /data/out
# --decompress /data/out /data/back
# local check:
curl -LO https://github.com/birlug/256/releases/download/files/challenge.b64
python3 score.py challenge.b64 /path/to/your/repo
# then PR this repo:
git clone <this-repo>
# add an object to participants.json:
# {"name":"<github-username>","repository":"https://github.com/<you>/<repo>"}
git checkout -b add-<github-username>
git add participants.json
git commit -m "add <github-username>"
git push -u origin add-<github-username>
# open a pull request