OpenAI

GPT-5.6-Sol

Verified Model id gpt-5.6-sol, run at each effort level it offers by us and by people we invited. Runs of the same effort are averaged.

Best Fast33.5at max effort
Best Thorough–not run yet
Suites run2
Our notes · 17 Oct 2026

GPT-5.6-Sol cleared the 2×2 but stopped early on 3×3

GPT-5.6-Sol sits 7th on the Index; at max, it cleared the 2×2 ladder without a miss but gave up on the one-shot 3×3 with 2 lives left.

Where it stands

GPT-5.6-Sol scores 33.5 at max, placing it 7th among 16 models on our Index. Just above it is Claude Opus 5 at max, with 35.0. Both cleared all 11 levels of the 2×2 ladder and reached level 5 of 20 on the one-shot 3×3. The difference was the interactive 3×3, where Opus 5 reached level 5 and GPT-5.6-Sol reached level 4.

We ran high and max. Raising effort added 8 completed levels on the 2×2, but only 1 on each 3×3 test. Neither effort unlocked the 4×4 Gauntlet.

What stood out

At max, GPT-5.6-Sol gave up on level 6 of the one-shot 3×3, where it had to submit one complete answer without seeing the result first. It still had 2 lives left. An earlier miss at level 1 was its only failed answer before it stopped; it had then solved through level 5. The round lasted 45 min 54 s and used 101,420 output tokens, more than either of its other max rounds. At high, the same test had ended in 6 min 56 s after misses at level 1 and twice at level 5. The max run went 1 level further but stopped without exhausting its chances.

The 3×3 cube at level 6, as the model got it
Level 6, as the model got it. Replay

The max run’s 2×2 clearance included an 82-move solution to an 8-move scramble at level 8: 74 moves longer than the shortest solve we knew. In this interactive test, it could turn the cube and see each result. That single solution accounted for more than a quarter of the round’s 318 moves. The length did not cost it a life. It finished all 11 cubes without a miss, spending 26 min 31 s and $2.51 on the ladder. Compared with high, it completed 8 more levels for an additional 9 min 6 s.

move 0 / 86
Level 8, 82 moves, solved. Full replay

Effort by effort

  • high, 16.0: Reached level 3 on both interactive ladders and level 4 on the one-shot 3×3. Every round ended out of lives; all 3 misses on the 2×2 came at level 4.
  • max, 33.5: Cleared the 2×2, exhausted its lives on the interactive 3×3, and gave up on the one-shot test. The interactive 3×3 misses came once at level 4 and twice at level 5.
GPT-5.6-Sol, Fast v3 score at each effort.

Time and cost

The best effort, max, took 1 h 52 min 20 s at an API-equivalent cost of $8.56, compared with 35 min 11 s and $3.26 at high. The cheaper run did not come close to matching it: max solved 20 cubes across the suite against high’s 10. The largest expense at max was the interactive 3×3, which cost $4.60 and took 39 min 55 s. That round gained just 1 completed level over high, whose corresponding round cost $1.37 and lasted 10 min 50 s.

Fast v3 by effort

Three ladders: 2×2 and 3×3 interactive, and 3×3 one-shot. A perfect 100 on all three unlocks the 4×4 Gauntlet, two 4×4 ladders of 20 levels. No time limit; each ends at the top level or on the third miss.

EffortIndex2×2 ladder, interactive3×3 ladder, interactive3×3 ladder, one-shot4×4 Gauntlet, interactive4×4 Gauntlet, one-shotMovesTimeTokensCostRuns
high16.027.315.020.0––32235 min8.6M$3.261
max33.510020.025.0––5201h 52m23.6M$8.561
Every run (2)

What it took

What the run cost, split into what the tokens were for: thinking, answers (moves and tool calls), fresh input, and input read back from cache. Sessions its API cut off are left out. For the whole suite.

  • Thinking
  • Answers
  • Input
  • Cached input

Split at the model’s list prices, scaled to what the run cost. Many runs themselves used subscriptions.

Level by level

The moves each solve took at each level of a ladder. Level N is scrambled N moves from solved, so the diagonal is the scramble undone move by move. A line that jumps off it has switched to a method of its own: longer, but it still works when the scramble is too deep to see.

Compare with other models · The Index · All boards