SUDOKU BENCHMARK ยท COZY

gpt-oss 20B on Cozy grid 1

Changed a given digit: 46 of 50 blank cells right, 4 given digits changed. It said it was 100% sure. All of gpt-oss 20B's answers.

RESULTChanged a given digit
CELLS RIGHT46 / 50
SAID100% sure
MODEL TIME12.6 min
COSTlocal, no list price
TOKENS IN / OUT342 / 8,212

5 given 5 filled correctly 53 wrong, with the right digit 3 left empty

The end of its reply

<grid>
576243189
493718652
281569743
658492317
127385946
349716528
762951384
914837625
835624971
</grid>
<confidence>100</confidence>

Grid cozy-0 of set v1, 31 givens. Puzzle as one line: 5..........3..8..2.8..6974..5.49..1.12.....96.49.76..8.6295..3.9..8..2..........1