GPT-6 Luna

33.5± 7.5 (95% CI)

mini-SWE · pass rate 71.4% · simplicity 62.8

Runs

Each cell is one full attempt: GPT-6 Luna solves every problem with the task's production library. Click a cell to open that run.

TaskRun 1Run 2Run 3Mean ± std

arazu 27.3 27.1 28.127.5 ± 0.5

clirs 33.7 29.4 36.633.2 ± 3.6

heretical 33.3 26.4 33.231.0 ± 4.0

npr 36.1 39.7 27.834.5 ± 6.1

plaid 6.1 20.2 24.917.1 ± 9.7

pyda 58.6 58.2 53.256.7 ± 3.0

canon 30.6 36.6 36.434.5 ± 3.4

roadkill 7.3 6.1 4.86.1 ± 1.3

rubberband 45.4 41.5 42.243.1 ± 2.1

weft 22.7 26.7 28.425.9 ± 3.0

sapi 25.4 32.1 29.729.1 ± 3.4

simu 27.1 25.3 32.128.2 ± 3.6

uglypie 58.7 52.3 58.956.6 ± 3.8

umami 48.9 55.0 48.950.9 ± 3.5

wapi 15.4 15.8 5.912.4 ± 5.6