Verified Run by us through Codex, locked to the cube tools. Finished 23 Oct 2026 21:05 UTC.
GPT-5.6-Luna scores 8.1 at low, below its best max run at 20.8. It gave up on the interactive 2×2 ladder at level 3 with all 3 lives left. It had solved the first 2 cubes without a miss, then ended the round rather than using any of its remaining lives.
The 2×2 round ended after 24 s and 16 moves across its 3 cubes. On the interactive 3×3, it again solved the first 2 levels, then gave up at level 3 with 3 lives left, after 32 s. The one-shot round ended differently: two wrong answers at level 1 used up two lives before it solved a cube there. A wrong answer at level 2 cost the final life. Across the suite, we saw 5 solved cubes; the Gauntlet stayed locked.
The suite took 75 s and cost $0.03, the fastest and cheapest of its runs. The other low run took 148 s and cost $0.05 for a lower score. Its best max result required 3 h 37 min 30 s and $1.72.
At low, it solved through level 2 of 11, then gave up on level 3 with all 3 lives left. Half the round’s 24 s went on level 2, solved in 6 moves, 4 more than the scramble. On the final cube, it used 9 of the allowed 36 moves across 2 calls before stopping, without exhausting either its moves or its lives.
At low, it reached level 2 of 20, then gave up on level 3 without making a move, with all 3 lives left. Level 1 took 1 move and 2 s; level 2 needed 18 moves, 16 more than the scramble, across 7 calls. That second cube accounted for 28 s of the 32 s round, which ended without a miss.
It solved level 1 of 20, then lost its last life on level 2. The first two answers failed on single-move scrambles: D' for D2, then U' for U'; only the third attempt, F2, worked. The final cube took 10 s, longer than any earlier attempt, but its answer R D2 did not undo the scramble L' U2.