Verified Run by Rohit @rohit3a through Claude Code, locked to the cube tools. Finished 12 Oct 2026 03:27 UTC.
Claude Opus 5 scores 35.0 at max, its best result and the one that places it 6th of 16 on the Index. It cleared the 2×2 ladder, but both 3×3 rounds ended when it ran out of output. The Gauntlet stayed locked.
The 2×2 run finished all 11 levels without a miss, using 117 moves and 601,304 thinking tokens. On the 3×3 interactive ladder, it completed level 5, then never answered level 6. The one-shot round ended at the same point. Each unanswered cube consumed about 511,700 thinking tokens across 4 consecutive replies that exceeded the model’s output limit: 96 min on the interactive cube and 95 min on the one-shot cube. Both tests ended with all 3 lives unused, keeping the levels already solved but earning nothing for level 6.
We ran only max, so there is no lower-effort result to compare. The suite took 6 h 43 min and cost $70.57 at API-equivalent rates. The successful 2×2 round accounted for 2 h 38 min and $30.31; the two unanswered 3×3 cubes alone occupied more than 3 h.
Cleared all 11 levels with all 3 lives intact and no misses. The first 5 solutions matched their scramble lengths, but level 8 took 1 h 11 min of the 2 h 38 min round: it hit the output limit 4 times before solving in 18 moves. The remaining cubes took less time but still needed longer solutions, including 29 moves for the 9-move scramble and 23 moves for the final 11-move scramble.
At max, it reached level 5 of 20 before running out of output on level 6, ending with all 3 lives unused. It never answered that cube: 4 consecutive replies exceeded its output limit, using about 511,700 thinking tokens over 96 min. The first 5 cubes all fell to exact inverses of their scrambles, with no misses; the unanswered cube took longer than all those solves combined.
At max, it reached level 5 of 20 before running out of output on level 6, leaving all 3 lives unused. Four consecutive replies hit its output limit over 95 min, totaling about 511,700 thinking tokens without an answer. The first five cubes all fell to exact inverses of their scrambles, though level 5 alone took nearly 15 min and 76,300 thinking tokens.