GPT-5.6 Terra

40.0± 6.6 (95% CI)

mini-SWE · pass rate 82.2% · simplicity 57.7

Runs

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

TaskRun 1Run 2Run 3Mean ± std

arazu 32.9 35.1 36.834.9 ± 2.0

clirs 41.9 45.8 44.544.1 ± 2.0

heretical 33.5 30.2 34.032.6 ± 2.0

npr 46.3 49.0 53.449.6 ± 3.6

plaid 53.9 43.7 35.844.5 ± 9.1

pyda 59.8 45.9 52.852.9 ± 7.0

canon 39.9 42.1 39.640.6 ± 1.4

roadkill 14.1 13.0 9.512.2 ± 2.4

rubberband 49.6 53.3 52.051.7 ± 1.9

weft 26.6 31.2 19.325.7 ± 6.0

sapi 34.3 31.8 32.732.9 ± 1.3

simu 34.6 31.3 40.535.5 ± 4.6

uglypie 57.1 50.6 59.055.6 ± 4.4

umami 53.0 58.6 52.754.7 ± 3.3

wapi 22.3 11.7 16.216.7 ± 5.4