Verified Run by Rohit @rohit3a through Claude Code, locked to the cube tools. Finished 12 Oct 2026 16:28 UTC.
Claude Opus 5.5 scores 88.5 at xhigh, its best result, placing it 1st of 16 on the Index. It went further than every other model on both 4×4 Gauntlet ladders: level 8 of 20 interactive and level 9 of 20 one-shot.
It cleared the 2×2 ladder and the 3×3 interactive ladder without a miss, then cleared 3×3 one-shot, including a 123-move solution to a 12-move scramble. In the interactive Gauntlet, it deliberately exhausted the move limits on 2 cubes to get fresh ones. The failed level-8 attempt consumed 747,650 thinking tokens and 111 min before it finished by repeating U 93 times. It later gave up on level 9 with 1 life left. The one-shot Gauntlet ended with another give-up, on level 10 with 1 life remaining, after it deliberately failed 2 level-9 cubes and solved the third in 15 moves.
We measured 12 h 44 min and $156.40 in API-equivalent cost for xhigh. The medium run came close to its result in 7 h 24 min for $101.39. At max, it scored lower despite taking 13 h and costing $155.92.
Cleared all 11 levels on the first try, keeping all 3 lives. The first 6 solutions matched their scramble lengths, but level 7 took 26 min 49 s, more than half the round, and used 169,206 thinking tokens before finishing in 21 moves, 14 more than the scramble. Levels 8 through 10 also needed longer solutions; the final cube was solved in 11 moves in 77 s.
It cleared all 20 levels with all 3 lives left. The first 7 solutions exactly reversed the scrambles, but level 6 alone took 14 min 21 s. At level 9 it announced a slot-by-slot method and took 15 min 45 s, the longest cube of the round. The later solves used more moves but less time: level 20 finished in 77 moves and 2 min 56 s.
Cleared all 20 levels at xhigh with all 3 lives left. Level 12 took 123 moves against a 12-move scramble, the longest solution of the round. The first 8 answers matched their scramble lengths; every solution after that used between 72 and 123 moves, without a miss. Levels 8 and 9 together took about 2 h 15 min, more than half the round’s time, before returning solutions of 8 and 89 moves.
It reached level 8 of 20 at xhigh, one level beyond GPT-6-Astra at max, then gave up on level 9 with 1 life left. It deliberately exhausted the move limits on cubes at levels 8 and 9 to get fresh ones, repeating U 93 times on the first. That failed level-8 attempt took 111 min and 747,650 thinking tokens, over half the round’s time. It said it stopped the final cube to avoid adding wasted moves to the tie-breaker.
At xhigh, it reached level 9 of 20, then gave up on level 10 with 1 life left because it said it couldn't work out the solution. Level 8 alone took about 47 min and 334,950 thinking tokens for an 8-move solve. It deliberately failed two level-9 cubes with a single U each, then solved the third in 15 moves, 6 more than the scramble. It finished one level ahead of GPT-6-Astra at max, the nearest model on this test.