Progress on the Hutter Prize for lossless compression. Participants submit algorithms to compress enwik9, a 1-gigabyte sample of Wikipedia text. Score is measured as the combined size of the compressed archive and the decompression program (lower is better).
The Hutter Prize rewards lossless compression of a fixed sample of Wikipedia text, a task closely tied to how well a program can predict text. While not a direct measure of AI’s impact, its record is a narrow indicator of software progress and is easy to verify.
Between the prize’s inauguration in July 2019 and May 2026, total progress amounted to a 5% improvement over the baseline. Rapid submissions in June and July 2026 tripled that to a 15% improvement.
We transcribe the enwik9 rows of the prize’s official table of current and previous records, using each record’s date and total size. The series starts with the pre-prize baseline from July 4, 2019, followed by the accepted prize records. The earlier 100-megabyte enwik8 competition (2006–2017) is left out because its input is different, preventing direct comparison.
Source: the Hutter Prize.
Our charts, text and processing are available under the Creative Commons Attribution license. The underlying data comes from the Hutter Prize (https://www.hutter1.net/prize/index.htm) and is subject to its own licence or terms.
Progress on the Hutter Prize for lossless compression. Participants submit algorithms to compress enwik9, a 1-gigabyte sample of Wikipedia text. Score is measured as the combined size of the compressed archive and the decompression program (lower is better).