Verified Run by us through Codex, locked to the cube tools. Finished 26 Oct 2026 19:26 UTC.
GPT-6.1 Terra scores 14.5 at high, well below its best max result of 62.0. Its max result puts it 3rd of 18 models on the Index; high is not the effort it is ranked on. On the 2×2 interactive ladder, it solved 3 levels, then gave up on level 4 with all 3 lives left.
The 2×2 stop came after 38 s and 18 moves across the round. The 3×3 interactive ladder followed the same pattern: 3 levels solved, then a give up on level 4 with 3 lives remaining, after 52 s. One-shot reached the same level but ended differently, with 3 wrong answers on level 4 exhausting its lives in 40 s. Across the suite, it solved 9 cubes and never unlocked the Gauntlet.
We ran the suite in 2 min 10 s, compared with 61 s at low and 94 s at medium. The xhigh run lasted 10 min 9 s, while its best max run took 1 h 15 min 25 s. API-equivalent costs are unavailable for these efforts.
At high, it solved 3 of 11 levels, then gave up on level 4 with all 3 lives left. The first two cubes took 7 s combined, each solved in as many moves as the scramble; level 3 needed 6 moves and 14 s. It spent the remaining 17 s on level 4, stopping after 9 moves despite a 48-move allowance.
At high, it solved 3 of 20 levels, then gave up on level 4 with all 3 lives left. The first 3 cubes took 22 s, each using more moves than its scramble. Level 4 took the remaining 30 s: it made 14 moves across 7 calls before stopping, well short of the 48-move limit.
It reached level 3 of 20, then lost all 3 lives on wrong answers at level 4. The first 3 solutions exactly reversed their scrambles, with level 3 taking 15 s of the round’s 40 s. Each failed answer was also 4 moves long, matching the scramble’s length but leaving the cube unsolved.