OpenAI

GPT-6.1 Terra

Verified Model id gpt-6.1-terra, run at each effort level it offers by us. Runs of the same effort are averaged.

Best Fast62.0at max effort
Best Thorough–not run yet
Suites run7
Our notes · 26 Oct 2026

GPT-6.1 Terra cleared both interactive ladders at max

GPT-6.1 Terra sits 3rd on the Index, but across seven suites it gave up with lives left 10 times.

Where it stands

GPT-6.1 Terra ranks 3rd of 18 models on the Index, with its best effort, max, scoring 59.0. Just above it is GPT-6-Astra at max, on 87.5.

GPT-6.1 Terra’s Index result cleared all 11 levels of the 2×2 ladder and all 20 of the interactive 3×3, where it could see the cube after turning it. Its one-shot result, one complete answer per cube without seeing the outcome first, stopped after level 6 of 20. No GPT-6.1 Terra run unlocked the 4×4 Gauntlet. Both GPT-6-Astra and Claude Opus 5.5 cleared the one-shot ladder and reached it.

What stood out

We saw GPT-6.1 Terra give up with lives remaining in 10 rounds. Six were interactive rounds at low, medium and high, all abandoned with all 3 lives intact. Even at max, clearing both interactive ladders did not mean it would exhaust its chances on one-shot. One run stopped at level 7 with 3 lives left after spending 19 min 23 s on that round. It said: "Solved through level 6; gave up at level 7." Another max run reached level 8, missed at level 9, then gave up with 2 lives remaining.

The 3×3 cube at level 7, as the model got it
Level 7, as the model got it. Replay

At xhigh, GPT-6.1 Terra deliberately used up the move limit on two interactive 3×3 cubes at level 4 to get fresh cubes. It repeated R, a turn of the right face, 40 times on one of them. The two discarded cubes cost 2 lives. GPT-6.1 Terra then solved level 4, but gave up at level 5 with its last life still available. The whole round lasted 100 s and consumed 124 moves.

move 0 / 48
Level 4, 48 moves, not solved. Full replay

Effort by effort

  • low, 12.6: Reached level 2 on the 2×2 and level 3 on both 3×3 ladders. All three rounds ended by giving up, with only one wrong one-shot answer before stopping.
  • medium, 12.6: Matched low’s levels and score. The difference was one-shot: it submitted three unsuccessful answers at level 4 and ran out of lives rather than giving up.
  • high, 14.5: Added one solved 2×2 level, reaching level 3 on every ladder. Both interactive rounds ended at level 4 by giving up; three misses there ended one-shot.
  • xhigh, 30.5: Cleared the 2×2 with 2 lives left, using 435 moves across the round. The interactive 3×3 stopped after level 4; one-shot again lost every life at level 4.
  • max, three runs, 59.0, 62.0, 60.5: Cleared both interactive ladders every time; one-shot stopped after levels 6, 8 and 7, respectively. One interactive level-11 cube took 125 moves against an 11-move scramble.
GPT-6.1 Terra, Fast v3 score at each effort.

Time and cost

The max suite used for GPT-6.1 Terra’s Index result took 58 min 5 s. Across all seven suites, time ranged from 61 s at low to 1 h 15 min 25 s for the longest max run. API-equivalent costs are unavailable, so we cannot name a cheapest effort. No lower effort finished close to max: xhigh took 10 min 9 s but stopped 16 levels earlier on the interactive 3×3. Moving from low to medium added 33 s without adding a solved level.

Rewritten as new results came in. Earlier versions

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
low12.618.215.015.0––281 min389k$0.091
medium12.618.215.015.0––432 min538k$0.141
high14.527.315.015.0––622 min731k$0.161
xhigh30.510020.015.0––57710 min8.8M$0.951
max60.510010035.0––19631h 06m57.8M$5.273
Every run (7)

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. Some costs are estimated from tokens at list prices. 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