AIToday
Large Language ModelsArs Technica AIPublished: Oct 2, 2026, 04:00 JST

Ataraxos AI beats Pim Niemeijer 15-1 at Stratego

Ataraxos AI beats Pim Niemeijer 15-1 at Stratego

3 Key Points

  1. What happened

    Researchers from Carnegie Mellon, MIT, NYU and Stanford trained an AI called Ataraxos that beat Pim Niemeijer, arguably the best Stratego player ever, 15 games to one with four draws, using just 16 GPUs and a few thousand dollars.

  2. Why it matters

    The win came in a game where most information stays hidden, which earlier AI systems couldn't handle reliably — suggesting the method may work for other problems with long stretches of uncertainty.

  3. What to watch

    The result hinges on whether one match against one player proves the approach generalizes. Watch whether the team releases more matches or faces other top players.

WHO IT HITSAI researchers and game-theory teams working on imperfect-information problems may take note, since the training cost was small enough to be replicated outside big labs. Tournament organizers for strategy games could also see elite human play challenged in ways previously seen in chess and Go.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

Stratego had long resisted computer mastery even as chess, Go, and poker fell to machines. Deep Blue beat Garry Kasparov in 1997, AlphaGo beat Lee Sedol in 2016, and poker bots have beaten professionals for years. DeepMind, despite its exceptional budget, couldn't build a machine to reliably beat the best Stratego players.

The challenge came from the game's structure. In Stratego, each player gets 40 pieces, and only collisions reveal identities. MIT computer scientist Gabriele Farina noted that Texas Hold'em has just 1,326 possible hands, but Stratego has more than a decillion possible setups. Games can easily last 2,000 moves, compared to about 40 in chess. Bluffing adds another layer: move a weak piece like a marshal to scare the opponent, but bluff too often and threats lose meaning.

NYU researcher Eugene Vinitsky pointed to the distinctive scale of hidden information unfolding over a long time. The win with 16 GPUs suggests smaller teams may now tackle similar problems, though whether this approach extends beyond Stratego remains to be seen.

FAQ
How much did it cost to train Ataraxos?
It took just 16 GPUs and a few thousand dollars to train.
Who built Ataraxos?
A team of researchers from Carnegie Mellon, MIT, New York University, and Stanford University.
What was the final score of the match?
Ataraxos beat Pim Niemeijer 15 games to one, with four draws.
Ars Technica AIRead Original Article

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleJudge Amit Mehta dismisses Chegg, PMC antitrust suits over AI Overviews