Verified Run by Rohit @rohit3a through Codex, locked to the cube tools. Finished 11 Oct 2026 23:46 UTC.
GPT-5.6-Luna scored 6.3 at low, below its best max result of 20.8 and its other low run. It gave up on level 2 of the 2×2 interactive ladder with 2 lives left, after solving only the opening cube. A failed attempt had cost one life; it stopped rather than use the remaining tries.
The 2×2 round accounted for 54 s and 29 moves before that exit. On the 3×3 interactive ladder, it solved 2 cubes, then deliberately exhausted the move limit on all 3 level-3 attempts by repeating a single move, including R 26 times. Throwing cubes away to get fresh ones consumed every remaining life. The one-shot round ended with 3 wrong answers at level 2 after one successful solve. We saw 4 solves across the suite, and the Gauntlet stayed locked.
The run took 2 min 28 s and cost $0.05 in API-equivalent usage. Its other low run scored higher in 1 min 15 s for $0.03, nearly half the time. The medium run also scored higher, taking 3 min 32 s at $0.08.
At low, it solved level 1 of 11, then gave up on level 2 with 2 lives left. The opening cube took 1 move and 3 s; the first level 2 attempt used all 24 moves over 40 s without solving a cube scrambled by 2 moves. On the fresh cube, it made 4 moves in 11 s before calling give_up.
At low, it reached level 2 of 20 before losing all 3 lives on level 3. It threw away all 3 cubes there by repeating one move to exhaust the move limit and get a fresh cube, including R 26 times on the first. The opening solves took 3 and 10 moves against scrambles of 1 and 2 moves. Of the round’s 66 s, 32 s went on the first level 3 attempt.
It solved level 1 of 20, then lost all 3 lives on level 2 with incorrect answers. The opening D2 solved its cube in 3 s; each subsequent answer used 2 moves, matching the scramble’s length but leaving the cube unsolved. The first miss took 12 s, the longest attempt of the 28 s round.