Energy-Limited, Leak-Proof AI Benchmark Posts Its Biggest Gain in 20 Years as Trillions Ride on AI

[James Bowery]

For immediate releaseHutter Prize Committee

Energy-Limited, Leak-Proof AI Benchmark Posts Its Biggest Gain in 20 Years as Trillions Ride on AI

The Hutter Prize requires no validation data, so its scores cannot be inflated by data leaks. Three entries, each running on a single CPU core, just improved its record by nearly 10%.

September 30, 2026 — With trillions of dollars in data-center and power investment riding on claims of AI progress, the Hutter Prize for Lossless Compression of Human Knowledge today awarded the largest advance in its 20-year history on a benchmark that data leaks cannot inflate. Three entries together improved the record by nearly 10%, compressing a 1GB snapshot of Wikipedia by more than a factor of 10, each on one general-purpose CPU core.

Hutter Prize record compression factor, 2006–2026.¹ Source: Hutter Prize committee.

Most AI benchmarks score models on validation data: material held back from contestants so models cannot simply memorize it. If that validation data leaks into a model’s training data, by accident or otherwise, its score rises with no real gain in capability, and outsiders have no easy way to tell. The Hutter Prize requires no validation data. The full 1GB file is public, and entries are scored on how small they can make it, counting the program that rebuilds it. Anything a model memorizes has to be stored in that program and counts against its score.

The prize measures intelligence the way science tests a theory: by how well it predicts. Every bit a model fails to predict is a bit that must be stored, so the compressed file is a theory of human knowledge that pays for each of its mistakes.

“This year has seen the largest progress in its 20-year history,” said Marcus Hutter, the prize’s sponsor, author of the 2026 book Job-Less Utopia and advisor to Google DeepMind co-founder Shane Legg’s 2008 PhD thesis, “Machine Super Intelligence.”

Fixed compute, rising capability

Entries must run on a single general-purpose CPU core, within roughly 50 hours and under 10GB of RAM. Because the hardware budget is held constant, every gain on the leaderboard comes from better algorithms rather than more chips or more electricity. The result is a counterpoint to the assumption that progress in machine intelligence must track growth in data-center power demand. The committee sets these limits for three reasons:

1. Specialized hardware biases research toward the industry’s current assumptions.
2. Talented individuals, not only well-funded labs, can afford to compete.
3. The public increasingly opposes the resource consumption of superintelligence efforts.

The committee likens resource-limited compression to decision accuracy under a reaction-time limit. Unlike bell-curve measures of intelligence, it sits on an absolute scale, and rote memorization is penalized by its size.

Next-token prediction, a decade early

The prize launched in 2006 on the premise that predicting text is the core of intelligence, anticipating 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 checks only whether a machine can fool a human, while the Hutter Prize tests for superintelligence.

The winners

• Ibrahim Marcouch and Kaido Oravcmix-lex1% better
• David Freelancmix-obias+1% better still
• Vladimir Ivanov (new record holder)fx2-cmix-T+7.5% better still

Each improvement is measured against the record set by the entry before it. 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, which caught the judging committee by surprise; it is now racing to use the same tools 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 for every 1% improvement, with a minimum improvement of 1% over the current record. Rules, records and past winners are at prize.hutter1.net.

About the Hutter Prize

The Hutter Prize rewards advances in lossless compression of human knowledge as a rigorous measure of superintelligence. It launched in 2006 at €50,000 scale on enwik8, a 100MB Wikipedia snapshot, and scored entries on compression ratio alone. In 2020 it expanded to €500,000 scale, moved to the 1GB enwik9 snapshot, and began counting the size of the compressor along with the compressed data. The committee is Marcus Hutter (sponsor and arbiter), Matt Mahoney (competition operations) and Jim Bowery (claim verification and public relations).

Media contact: Jim Bowery, Hutter Prize committee
jabo...@gmail.com
prize.hutter1.net

###

¹ In 2020 the prize moved from the 100MB enwik8 file to the 1GB enwik9 file and expanded from €50,000 to €500,000 scale. Earlier enwik8 records are normalized to 1GB so both eras share one scale. The enwik9 metric includes the length of the compressor. ↩

[James Bowery]

For a web page of the press release unconstrained by Google Groups' formatting issues;

https://spasim.org/docs/HutterPrize/TwentyYearBenchmarkShattered.html

--
You received this message because you are subscribed to the Google Groups "Hutter Prize" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hutter-prize...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hutter-prize/8d9211bc-9a56-498a-95b4-a8158b9191d9n%40googlegroups.com.

[surya milenial]

Dear all

Does the hutter prize still continued after the winner announcement? 

To view this discussion visit https://groups.google.com/d/msgid/hutter-prize/CAN%3DDHybNTpdB2bj6_V%3DyQY1zM%3DZKmorPQf85dd0ayrrX-%3DYsWQ%40mail.gmail.com.

[Matt Mahoney]

Yes, the Hutter prize is still open. There are currently 3 more pending entries plus 4 unqualified that exceed the current leader, cmix-lex-transformer. You can find them here.

https://mattmahoney.net/dc/text.html

Cmix-neif 1.0 has passed all our tests and is open for public comment. It improves on the current leader cmix-lex-transformer by over 4%.

Zmix and cmix-lex-transformer lexth11c each improve on the leader by over 1%. In my tests, zmix decompression was over the time limit. It would still qualify if it runs fast enough on another machine. Lexth11c decompression was fast enough but I did not test compression.

There are also 4 other programs listed on LTCB but all of them exceed resource limits or improve by less than 1% over earlier programs and are not eligible.

Sizes shown below are compressor size plus self extracting archive size. All are open for public comment.

July 24, fx2-cmix-transformer, 100,424,672, awarded.

Sept 14, zmix, 99,312,424, decompression over time, pending.

Sept 16, cmix-neif-pre2, 98,249,835, decompression over time, pending.

Sept 20, cmix-neif-pre3, 97,261,675, decompression over time, pending.

Sept 27, cmix-lex-transformer lexth11c, 99,313,750, decompression passed, pending.

Sept 28, cmix-neif-pre3f, 97,262,944, decompression over time, pending.

Oct 5, cmix-neif-1.0, 95,914,051, passed all tests.

I did not test compression on any of these. zmix is eligible if it passes on another machine. cmix-lex-transformer lexth11c is eligible only if compression passes and zmix, pre2, and pre3 all fail.

-- Matt Mahoney, mattma...@gmail.com

To view this discussion visit https://groups.google.com/d/msgid/hutter-prize/CANYu7B6cXQYZQNPHkebZ0Qb5qEhyoy5dafj_gFGmHQasgXLMMQ%40mail.gmail.com.