Verified Run by Hars @thisishars through Codex, locked to the cube tools. Finished 11 Oct 2026 23:59 UTC.
GPT-6-Luna scores 14.5 at max, matching its best xhigh result and placing it 17th on the Index. It gave up on the interactive 3×3 at level 3 with 2 lives left. Its explanation was specific: "Reached level 2; level 3 remains active, but Cube Arena rejected the displayed attempt ID, so I couldn’t finish the test."
The 2×2 ladder ended with all 3 lives lost at level 4, after 160 moves across 6 cubes. On the interactive 3×3, one level 3 miss preceded the decision to give up. The one-shot 3×3 went furthest: it recovered from a miss at level 2, reached level 4, then lost its remaining lives at level 5. None of the ladders cleared, so the Gauntlet stayed locked.
We recorded 53 min 11 s and an API-equivalent cost of $0.24 for the suite. The one-shot round accounted for 30 min 43 s, more than the two interactive rounds combined. The matching xhigh run finished sooner; the other max runs cost between $0.23 and $0.35 and all scored lower.
Reached level 3 of 11, then lost all 3 lives at level 4 by exhausting the 48-move limit each time. The first miss took 9 min 2 s, more than half the round; the next two used another 96 moves in just 23 s combined. Before that, it solved the opening cube in 1 move, but needed 6 and 9 moves for cubes scrambled by 2 and 3 moves respectively.
At max, it reached level 2 of 20, then gave up on level 3 with 2 lives left. The first two solutions exactly reversed their scrambles; the first level 3 attempt took 6 min 33 s and exhausted all 36 moves. On the fresh cube, it made no moves before stopping, saying Cube Arena had rejected the displayed attempt ID.
Reached level 4 of 20, then lost its last 2 lives on wrong answers at level 5. The first miss came at level 2; all 4 successful answers exactly reversed their scrambles. Level 4 took the longest at 9 min 33 s for a 4-move solution, nearly a third of the round’s 30 min 43 s.