Start free trial

VERIFIED

A few thousand: the final run only

The "few thousand dollars" is the final training run, not the whole project.

“Ars Technica reports that Ataraxos, an AI from Carnegie Mellon, MIT, NYU and Stanford researchers, beat arguably the best Stratego player of all time 15 games to one with four draws, after training on 16 GPUs for a few thousand dollars.”

Ars Technica (Oct 1, 2026), reporting the Ataraxos team's paper

Recorded 3 October 2026. Our ruling means: we tried to break it and could not.

Read the story →

Receipts pack

The evidence file for this verdict

16 receipts: 9 support the claim, 1 contradict it, 6 add context. 4 figures on the record, 2 traced to a quoted receipt.

The receipts

Each source, the line we quote from it, its date and which way it cuts. Free to read and to cite.

  1. 1Supports the claimarstechnica.com ↗

    player of all time, 15 games to one, with four draws

    Look up an archived copy ↗

  2. 2Supports the claimarstechnica.com ↗

    And it took just 16 GPUs and a few thousand dollars to train it.

    Look up an archived copy ↗

  3. 3Contextarstechnica.com ↗

    That balancing act, the team explains, is what stumped earlier AIs like DeepMind’s DeepNash, introduced in 2022.

    Look up an archived copy ↗

  4. 4Supports the claimarxiv.org ↗

    In a 20-game series, Ataraxos defeated the most decorated Stratego player of all time

    Look up an archived copy ↗

  5. 5Supports the claimarxiv.org ↗

    Ataraxos won the series with 15 wins, 1 loss, and 4 draws

    Look up an archived copy ↗

  6. 6Supports the claimarxiv.org ↗

    over 600 weeks as the #1 ranked player (the most of all time)

    Look up an archived copy ↗

  7. 7Supports the claimarxiv.org ↗

    Such a run costs less than $8,000 at current prices

    Look up an archived copy ↗

  8. 8Contextarxiv.org ↗

    such a training run would roughly cost between $3,000,000 and $4,500,000

    Look up an archived copy ↗

  9. 9Contextarxiv.org ↗

    To the recollection of the corresponding author with whom we spoke

    Look up an archived copy ↗

  10. 10Contextarxiv.org ↗

    DeepMind responded that it would not be possible as the code for DeepNash is no longer functional.

    Look up an archived copy ↗

  11. 11Contextarxiv.org ↗

    the outcomes of the games were far from independently and identically distributed

    Look up an archived copy ↗

  12. 12Contextarxiv.org ↗

    Pim was paid $1,000 for participating in the evaluation

    Look up an archived copy ↗

  13. 13Supports the claimarxiv.org ↗

    Across 40 such games, Ataraxos recorded a vertiginously high 95% effective win rate (38 wins, 2 losses, 0 draws)

    Look up an archived copy ↗

  14. 14Supports the claimarxiv.org ↗

    doing so requires not an industrial budget, but merely a few thousand dollars

    Look up an archived copy ↗

  15. 15Contradicts the claimnews.ycombinator.com ↗

    Just 16 GPUs, and a few thousand dollars?

    Look up an archived copy ↗

  16. 16Supports the claimarxiv.org ↗

    Ataraxos achieved this result while only costing a few thousand dollars to train.

    Look up an archived copy ↗

Members' extras

What each receipt proves, and what would change our mind.

  • What each receipt above proves, in our words
  • The figures table, each number traced to the receipt that states it
  • What would change our verdict
  • Our call on the record, with its due date

The verdict, the opening and every receipt above stay free. Membership adds our analysis to the evidence file for every checked story.

First membership: 30 days free, then A$89 a year. One introductory trial per customer. Card required; renews annually until cancelled. Cancel before the trial ends to avoid the first charge.