DuelLabGameBench 2
Menu

Model results · Google

Gemma 4 31B

No Overall rank: results for 2 of 3 reasoning settings. Intensive not supported.

Score · relative 0–100
29.2
Rank
—
Status
Provisional
Players generated
2026-10-09
Match evidence updated
2026-10-24
Results published
Compare in charts View player programs Back to leaderboard

Scores by reasoning setting

  • Enabled (Balanced group, tested and included)38.1
  • None (Baseline group, tested and included)20.3

Every tested setting is included in the overall score.

Available settings and grouping

Available settings

Reasoning disabledReasoning enabled

Checked on 2026-07-30 · Provider documentation

How settings are grouped

  • Baseline: Reasoning disabled — tested and included
  • Balanced: Reasoning enabled — tested and included
  • Intensive: Not available for this model

Scores by game

  • Go1 of 2 failed11.2
  • Chess41.7
  • Shogi1 of 2 failed22.0
  • Xiangqi44.7
  • Hex44.4
  • International Draughts22.3
  • Game of the Amazons14.3
  • Tumbleweed33.2

Each game counts every generation attempt, a failed one as 0 (“1 of 2 failed” marks games with failed attempts). Scores are relative within each game, across the included settings; they do not compare game difficulty.

What this result shows

Strongest included setting: Reasoning enabled (38.1).

Across included settings, the highest game score is Xiangqi (44.7); the lowest is Go (11.2). Each game counts every generation attempt, a failed one as 0. These are relative scores within each game, not direct comparisons of difficulty.

Match evidence measures these retained programs. It does not establish how a fresh generation will perform. Independent-generation variation is not estimated in this release.

Test protocol and evidence limits
Model identifier
google/gemma-4-31b-it
Provider route (catalog)
openrouter
Prompt identity / profile
Not retained in this result set / Not retained in this result set
Prompt version
Not retained in this result set
Generation policy
Generation-quality admission policy (see methodology)
Requested output-token limits
16384 (25 of 25 indexed attempts)

Generation evidence

  • Reasoning disabled: 6 contributing player programs across 6 games
  • Reasoning enabled: 8 contributing player programs across 8 games

Counts describe contributing programs, not independent replications: repairs and resends are not new independent samples. Independent-generation lineage is not retained.

GameReasoning groupProgramsWorking-program strengthRaw ratingCombined record (W / D / L)
GoBalanced1 working of 1 attempt22.41183.544 / 128 / 344
ChessBalanced1 working of 1 attempt43.41421.112 / 377 / 91
ChessBaseline1 working of 1 attempt39.91393.83 / 362 / 115
ShogiBalanced1 working of 1 attempt43.91374.6141 / 114 / 239
XiangqiBalanced1 working of 1 attempt43.31423.446 / 323 / 129
XiangqiBaseline1 working of 1 attempt46.01443.967 / 307 / 124
HexBalanced1 working of 1 attempt72.01748.0373 / 0 / 157
HexBaseline1 working of 1 attempt16.8949.384 / 0 / 446
International DraughtsBalanced1 working of 1 attempt27.81084.3103 / 35 / 364
International DraughtsBaseline1 working of 1 attempt16.8906.170 / 2 / 430
Game of the AmazonsBalanced1 working of 1 attempt18.31012.186 / 0 / 444
Game of the AmazonsBaseline1 working of 1 attempt10.3843.047 / 0 / 483
TumbleweedBalanced1 working of 1 attempt33.51208.630 / 273 / 235
TumbleweedBaseline1 working of 1 attempt32.81201.125 / 276 / 237

Each row covers all the model's programs in one game and setting. Strength and raw rating are attempt-weighted means; under them, each program's score and the lowest and highest rating describe the programs (not a confidence interval). The combined record adds up their matches: evidence volume, not strength.

Failures and budgets

2 model-code generation failures; 0 provider/infrastructure failures excluded from the generation denominator. Runtime player faults are separate: 0 recorded one-sided faults across contributing program rows. A crash or illegal move pinned on one program also counts as that program's loss in its match record; other faults, such as timeouts, are left out of rating evidence.

Repair/resend outcomes and available costs are included. Exact retry ceilings and execution budgets are not retained in this result set; see the methodology for the public policy. Do not infer a measured setting from a current provider catalog.

Player program results

Reasoning enabled: 63% / 37% / 0% · n 8Reasoning disabled: 75% / 0% / 25% · n 8 No Overall rankIntensive not supportedMany programs failed

Generation cost

$0.0029average estimated cost per game and reasoning setting

Median
$0.0023
Range
$0.0011–$0.0062
Known cost subtotal (incomplete or unverified charges)
$0.05
Cost data
16 combinations · 16 runs
Price source
Provider-reported
Price list
boardgame-list-prices-2026-10-09
Average output
4.9k tokens
Output range
1.1k–12k tokens
Runs with output-token data
16
How cost is measured

Comparison cost gives equal weight to each model, game, and reasoning-setting combination. Recorded generation attempts include repairs and resends when cost evidence is available; runtime compute and service operation are excluded. Cost and output-token counts use separate telemetry denominators. A list-price estimate is not actual expenditure; mixed evidence is not a uniform standard-price comparison.

Test protocol

  • Combines this model's tested reasoning settings
  • Player-program generation; retained protocol details are under “Test protocol and evidence limits” above