Hutter Prize: compress enwik9 below 110,793,128 bytes

Losslessly compress enwik9, the first 10^9 bytes of English Wikipedia, smaller than anyone has — counting the decompressor itself as part of the output.

Bar to beat. S = S_1 + S_2 < 110{,}793{,}128 bytes, where S_1 is the size of the decompressor and S_2 the size of the archive. The prize pays 500{,}000 \times (1 - S/L) euros against that limit L, with a minimum claim of 1% (5,000 €).

Submission format. A self-contained compressor and decompressor, plus the archive they produce.

Verifier sketch. Archetype D, a simulator oracle: run the submitted decompressor in the sandbox, byte-compare its output against the pinned enwik9 (known SHA-256), and sum the two sizes. The official rules also cap runtime and memory on a reference machine; a verifier has to adopt them verbatim or state its own.

Not written yet because it executes an untrusted binary for hours against a 1 GB file. That is a different resource class from every other verifier here (10 s, 256 MB) and needs its own sandbox policy, timeout and storage budget.

Known. The record has moved in small steps since 2006 and the contest is open-ended, with eight winners so far. All leading entries are context-mixing models in the PAQ/cmix lineage.

Why. Hutter's premise is that compressing text well requires understanding it, which makes this the one public benchmark treating compression as an intelligence measure. The engineering transfers directly to archival storage.

Ref: Hutter Prize · Wikipedia: Hutter Prize