Hutter Prize Awards Largest Advance in 20 Years on Its Superintelligence Benchmark
Three entries slashed the knowledge compression record for 1GB of Wikipedia on a single CPU, under a test where every error penalizes.
September 30, 2026: The Hutter Prize for Lossless Compression of Human Knowledge today awarded three entries that together improved its record by nearly 10%, by far the largest single-year advance in the contest’s 20-year history. The prize measures intelligence the way science tests a theory: by how well it predicts. Every bit a theory fails to predict is a bit that must be stored. What’s left is a theory of human knowledge that admits its mistakes. The new record compresses a 1GB snapshot of Wikipedia by over a factor of 10.
“This year has seen the largest progress in its 20[-year] history,” said Marcus Hutter, the prize’s sponsor, PhD advisor to Google DeepMind founders and author of the 2026 book “Job-Less Utopia”.
Next-token prediction, a decade early
The prize launched in 2006 on the premise that predicting text is the core of intelligence. It anticipated today’s next-token prediction language models by more than a decade. The committee draws a sharp line between this benchmark and the Turing Test: the Turing Test only checks whether a machine can fool a human, while the Hutter Prize tests for superintelligence.
One CPU, by design
Entries must run on a single general-purpose CPU core, within roughly 50 hours and under 10GB of RAM. The committee sets these limits for three reasons:
- Specialized hardware biases research toward the industry’s current assumptions.
- Talented individuals, not only well-funded labs, can afford to compete.
- The public increasingly opposes the resource consumption of superintelligence efforts.
Resource limited compression is similar to decision reaction time accuracy. Unlike bell curve intelligence, it is on an absolute scale with rote memorization penalized by size.
The winners
- Ibrahim Marcouch and Kaido Orav, for cmix-lex 1% better
- David Freelan, for cmix-obias 1% better still
- Vladimir Ivanov, for fx2-cmix-T, the new record holder 7.5% better still
The three awards total nearly €50,000. Judging took longer than usual because the rate of new entries has exploded. Contestants are now using superintelligence to improve their machine learning algorithms. As a result the judging committee was caught by surprise and is in a race to use superintelligence to automate judging. This is feasible because the award criteria were designed to leave little discretion to the judges.
The next target
Awards are 1% of €500,000 per 1% improvement, with a minimum of 1% improvement. Rules, records and past winners are at http://prize.hutter1.net/.
About the Hutter Prize
Initially launched in 2006 with €50,000 scale, and expanded to €500,000 in 2020, the Hutter Prize rewards advances in lossless compression of human knowledge as a rigorous measure of superintelligence. The committee is Marcus Hutter (sponsor and arbiter), Matt Mahoney (competition operations) and Jim Bowery (claim verification and public relations).