SUDOKU BENCHMARK · LOCAL, THINKING LOW

Qwen3.8 Flash-Next

Alibaba · open weights · released 2026-08-26 · route qwen3.8-flash-next · Model profile · Chess results

Results

Score
5295%: 37 to 68 · 21 of 42 grids so far
Grids solved
11 of 21
Blank cells filled correctly
51.0%
Wrong cells / empty cells / changed a given / no grid
0 / 0 / 0 / 10
Broke the rules (complete grid repeating a digit)
0
Near misses (three cells or fewer wrong)
0
Average stated confidence
100%
Average confidence when it had not solved it
n/a
Sure but wrong (90% or more, unsolved)
0
Calibration (Brier, lower is better)
0.000
API-equivalent cost per grid
n/a
API-equivalent cost per solved grid
n/a
Seconds per grid
3,414
Tokens per grid (in / out)
605 / 64,273
Reasoning tokens per grid
n/a
Retries for an unreadable reply
13
Aborted on our side (not counted)
0

By tier

TierNeedsSolvedOutput tokensSeconds
CozySingles only2 of 337,3211,760
GentleScanning3 of 336,5211,762
SteadyLocked candidates1 of 391,2604,621
TrickyHidden pairs and triples1 of 372,3414,932
DeviousWings and fish2 of 354,0112,694
ExpertDeep wings and fish2 of 364,0973,240
BeyondNo human technique finishes it0 of 394,3614,893

Cells by the technique they needed

Each blank cell is tagged with the simplest technique that fills it, given everything a solver could fill before it.

TechniqueCellsFilled correctly
Last free cell131100.0%
Scanning256100.0%
Locked candidates42100.0%
Hidden pairs and triples52100.0%
Wings and fish48100.0%
Deep wings and fish46100.0%

Every answer