SUDOKU BENCHMARK · REASONING MEDIUM

Gemini 4 Argon

Google · released 2026-09-30 · route gemini-4-argon · Model profile · Chess results

Results

Score
2095%: 0 to 44 · 5 of 42 grids so far
Grids solved
1 of 5
Blank cells filled correctly
46.5%
Wrong cells / empty cells / changed a given / no grid
4 / 0 / 0 / 0
Broke the rules (complete grid repeating a digit)
4
Near misses (three cells or fewer wrong)
0
Average stated confidence
39%
Average confidence when it had not solved it
24%
Sure but wrong (90% or more, unsolved)
0
Calibration (Brier, lower is better)
0.053
API-equivalent cost per grid
$0.032
API-equivalent cost per solved grid
$0.162
Seconds per grid
42
Tokens per grid (in / out)
2,284 / 2,785
Reasoning tokens per grid
2,621
Retries for an unreadable reply
0
Aborted on our side (not counted)
0

By tier

TierNeedsSolvedOutput tokensSeconds
CozySingles only0 of 11108
GentleScanning0 of 137711
SteadyLocked candidates1 of 113,218169
TrickyHidden pairs and triples0 of 11108
DeviousWings and fish0 of 111012
ExpertDeep wings and fish0 of 0n/an/a
BeyondNo human technique finishes it0 of 0n/an/a

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 cell6155.7%
Scanning9933.3%
Locked candidates3797.3%
Hidden pairs and triples5232.7%
Wings and fish2025.0%

Every answer