GPT-5.6 Luna

30.1± 5.8 (95% CI)

mini-SWE · pass rate 79.0% · simplicity 46.5

Runs

Each cell is one full attempt: GPT-5.6 Luna solves every problem with the task's production library. Click a cell to open that run.

TaskRun 1Run 2Run 3Mean ± std

arazu 27.6 26.0 25.826.5 ± 1.0

clirs 37.3 35.1 39.137.2 ± 2.0

heretical 25.4 27.4 26.826.5 ± 1.0

npr 28.3 37.8 33.433.2 ± 4.7

plaid 34.1 24.4 28.629.0 ± 4.9

pyda 48.9 52.2 44.048.4 ± 4.2

canon 29.1 27.6 26.427.7 ± 1.4

roadkill 7.3 8.8 6.97.7 ± 1.0

rubberband 40.4 37.5 41.139.6 ± 1.9

weft 13.0 14.3 17.615.0 ± 2.3

sapi 25.8 30.3 22.226.1 ± 4.1

simu 27.1 30.0 23.326.8 ± 3.3

uglypie 43.9 45.4 42.944.0 ± 1.3

umami 37.1 38.4 41.439.0 ± 2.2

wapi 8.9 13.3 11.911.4 ± 2.2