GPT-5.6-Luna
OpenAI · max effort Verified
- Fast Index
- 20.0#30 of 64 · avg of 2
- Thorough
- –not run
- Coding
- 6.8#11 of 15
- Moves / solve
- 3112.5 cubes solved
- Cost (Fast)
- $1.532h 45m
Compare
Model vs model
Pick up to 4 models, at one effort or their best, and see them side by side: the Index and every test, level by level on the same cubes, moves, time, tokens, cost, how effort changes them, and the coding benchmark. Every chart and card downloads as an image.
Head to head
Every measure side by side; the best on each row is marked. Averages where an effort was run more than once.
| Measure | GPT-5.6-Luna (max) |
|---|---|
| Rows won | 0 |
| Fast suite | |
| Fast Index | 20.0 |
| Rank on the Fast Index | #30 of 64 |
| 4×4 Gauntlet | Gauntlet locked |
| 2×2 ladder, interactive | 36.4 · L4/11 |
| 3×3 ladder, interactive | 20.0 · L4/20 |
| 3×3 ladder, one-shot | 22.5 · L5/20 |
| Cubes solved | 12.5 |
| Moves on the shared cubes | – |
| Moves per cube solved | 31 |
| Time | 2h 45m |
| Input tokens | 16.9M |
| Output tokens | 810k |
| Cost | $1.53 |
| Cost per Index point | $0.08 |
| Minutes per Index point | 8.2 |
| Runs averaged | 2 |
| Score range over runs | 19.3–20.8 |
| Coding benchmark | |
| Coding score | 6.8 |
| Rank on the coding board | #11 of 15 |
| Hidden cubes solved | 50/50 |
| Moves per cube (solver) | 1035 |
| Solve time per cube | 6.5 s |
| Coding time used | 3h 21m |
| Coding cost | $3.04 |
Costs marked * are estimated from tokens at list prices. Ties share the mark.
Fast, test by test
Each test out of 100. The Index weighs the 2×2 ladder 20% and each 3×3 ladder 30%. A perfect 100 on all three unlocks the 4×4 Gauntlet: two 4×4 ladders of 20 levels, worth 10% each. Without it, 80 is the top score.
Level by level
Every level of each ladder, from each model's best run. The number is the moves it took; red marks are missed tries. Everyone gets the same cube on a level's first try, so the fewest moves there is starred. Click a level for its replay.
- solved first try
- solved after a miss
- missed, the ladder ended
- not reached
Does more effort help?
The Fast Index at every effort each model has been run at, whichever effort is picked above.
- GPT-5.6-Luna
Coding benchmark
Each model writes its own 5×5 solver in a few hours, then the frozen code solves hidden cubes. 100 = the 60-move reference; an unsolved cube scores 0.