Ataraxos's low-compute AI defeats the strongest human Stratego player

Researchers from Carnegie Mellon, MIT, NYU, and Stanford developed an AI called Ataraxos that defeated four-time world champion Pim Niemeijer—widely considered the greatest Stratego player ever—in Stratego, an imperfect-information game with heavy hidden information and long game timescales. According to a report by The Decoder on October 1, 2026, Ataraxos beat Niemeijer with an effective win rate of 85% over a three-week match series, and won 38 of 40 games in an exhibition match at the 2025 World Championship, using less than $8,000 worth of compute. A report by Ars Technica on October 2, however, said Ataraxos defeated Niemeijer 15 wins, 1 loss, and 4 draws, training on just 16 GPUs and a few thousand dollars. The two reports differ on specific figures: the earlier report cites an 85% effective win rate and 38 wins in 40 games (i.e., 38 wins and 2 losses), while the later report cites 15 wins, 1 loss, and 4 draws; on compute cost, the earlier report says under $8,000, while the later one says a few thousand dollars. Both reports note that Stratego's imperfect-information nature meant even DeepMind had previously failed to build an AI that could consistently beat top human players, so this low-compute achievement is seen as a breakthrough.

AI overview of reporting · Updated 17 days ago