Verified Run by Rohit @rohit3a through Cursor, locked to the cube tools. Finished 12 Oct 2026 13:28 UTC.
Composer 2.5 scores 6.3, its best result and the one that places it 18th of 16 on the Index. It solved 4 cubes across the suite. Neither interactive ladder got beyond level 1; one-shot went a level further. The Gauntlet stayed locked.
On the 2×2, it solved the opening cube in 1 move, then lost all 3 lives at level 2. The 3×3 interactive round ended the same way, with every miss on a cube scrambled by just 2 moves. Those rounds used 73 and 77 moves respectively, for one solve apiece. In 3×3 one-shot, where it submitted an answer without seeing the result, it recovered from a level 2 miss to solve a fresh cube, then lost its remaining lives at level 3.
We ran this suite in 31 min 27 s, with 26 min 55 s spent on the interactive rounds and 4 min 32 s on one-shot. There is no other effort run to compare, and no API-equivalent cost is available. The suite used 51,997 output tokens.
It reached level 1 of 11, then lost all 3 lives on level 2, exhausting the 24-move limit each time. The longest attempt took 7 min 15 s, more than half the round, cycling through half-turn pairs without solving a cube 2 moves from solved. On the first miss it said, "Wrong moves worsened the scramble; giving up to retry level 2 with a fresh cube."
It reached level 1 of 20, then lost all 3 lives at level 2 by exhausting the 24-move limit on each cube. The round took 13 min 20 s, with 10 min 11 s spent on those misses. Even the only solve took 5 moves against a 1-move scramble: it made 4 canceling turns of the back face before finishing with L'.
Reached level 2 of 20 before losing its last two lives to wrong answers at level 3. The first miss, at level 2, took 119 s of the round’s 4 min 32 s, longer than any other cube. Both successful answers exactly reversed their scrambles, but all three misses also matched the scramble length: the answers were short enough, just wrong.