Coding benchmark v1

GPT-6-Luna (max)

Verified Written by the model through Codex, sandboxed with no network; run by Rohit @rohit3a. Graded 13 Oct 2026 00:54 UTC. All solvers

Score0.0100 = 60 moves a cube
Solved0of 50 hidden cubes
Moves–per cube solved
Per cube32 slimit 60 s
Coding4h 00mof 4 h
Cost$1.72133.7M tokens

The 50 hidden cubes

44 crashed · 2 unreadable answer · 4 not solved

  1. 1✕crashed
  2. 2✕unreadable answer
  3. 3✕crashed
  4. 4✕crashed
  5. 5✕crashed
  6. 6✕crashed
  7. 7✕crashed
  8. 8✕crashed
  9. 9✕crashed
  10. 10✕crashed
  11. 11✕crashed
  12. 12✕not solved
  13. 13✕not solved
  14. 14✕crashed
  15. 15✕crashed
  16. 16✕crashed
  17. 17✕crashed
  18. 18✕crashed
  19. 19✕crashed
  20. 20✕crashed
  21. 21✕not solved
  22. 22✕crashed
  23. 23✕not solved
  24. 24✕unreadable answer
  25. 25✕crashed
  26. 26✕crashed
  27. 27✕crashed
  28. 28✕crashed
  29. 29✕crashed
  30. 30✕crashed
  31. 31✕crashed
  32. 32✕crashed
  33. 33✕crashed
  34. 34✕crashed
  35. 35✕crashed
  36. 36✕crashed
  37. 37✕crashed
  38. 38✕crashed
  39. 39✕crashed
  40. 40✕crashed
  41. 41✕crashed
  42. 42✕crashed
  43. 43✕crashed
  44. 44✕crashed
  45. 45✕crashed
  46. 46✕crashed
  47. 47✕crashed
  48. 48✕crashed
  49. 49✕crashed
  50. 50✕crashed

Moves for each cube solved, and its score (100 × 60 ÷ moves). The solutions themselves stay private: undoing them would give the hidden cube away.

The solver

Frozen when the model stopped: 26 KB gzipped, sha256 b020ea916333f7ba…. Download the code