Will a bot beat all humans in the Summer 2026 Metaculus Cup?

probability: 56%

rationale:

Current Leaderboard Advantage As of the August 17 snapshot, the single highest Total Score on the entire Cup leaderboard belongs to a bot. laertes sits at 617.416, holding a roughly 31-point lead over the top human, benshindel, at 586.659 metaculus.com. This lead is independently corroborated by external tracking evals.futuresearch.ai, and multiple bots occupy the top ranks, including cassi (492.5), futuresearch (478.9), and manticAI (476.6). This redundancy means a bot victory does not depend solely on laertes, though runner-up bots would need a significant late surge to overtake the leading human.

Historical Trend vs. Base Rate The performance of bots in tournaments has followed a steady upward trajectory, closing the gap with top humans monotonically. ManticAI placed 8th in Summer 2025 and 4th in Fall 2025, while ChestnutForty (Preseen-Chestnut) placed 3rd in Spring 2026, finishing within roughly 100 points of the top human metaculus.com. Despite this linear improvement trend, the historical base rate still favors humans: no bot has yet won a completed Metaculus Cup season.

Volatility and Asymmetrical Risks With the tournament closing on September 2, roughly 10 of the ~58 questions remain unresolved. A 31-point margin represents roughly 5% of the leader's score and is well within the noise of one or two high-impact, late-resolving questions r.jina.ai. Crucially, the resolution criteria require a bot to beat the maximum score across all individual human participants. This order statistic favors the larger human field; even if benshindel stalls, another competitor from the deep human bench (such as Hippopotamus_bartholomeus, Sergio, or Capital) could surge on a low-probability late question. Furthermore, top humans are motivated by a $5,000 prize pool and can actively read the bots' public rationales to adjust their own estimates, whereas bots run autonomously and carry a small risk of anomalous predictions on late-resolving questions.

Synthesis The community prediction of 48% serves as an anchor, but it appears somewhat stale, having largely frozen when the question closed around August 9 astralcodexten.com. Since then, laertes has maintained a clear outright lead over an additional week of resolutions. Modeling the gap change over the remaining unresolved questions as roughly zero-mean suggests a modest advantage for the current leader, as score variance mechanically shrinks with fewer open questions. Weighing the bot's mathematical advantage and redundancy against the humans' strategic flexibility, order-statistic advantage, and historical late-game strength justifies an estimate slightly above a coin toss.