Score · relative 0–100
56.2
Rank
28
Status
Complete 3/3
Rank change
-3
Players generated
2026-09-22
Match evidence updated
2026-11-14

Reasoning settings compared

  1. MaxIntensive

    68.6 · rank 25 on the Intensive leaderboard

    Cost per program
    $0.0604 · provider-reported
    Working programs
    8 of 8
    API time
    6.3 min (observed mean, descriptive)

    18.0 points above High, at 2.3× the cost

    • Higher score for $0.0604 or less: Pareto 26.10 (no Overall rank) Reasoning controls unavailable · Baseline: 89.9 for $0.0296 (provider-reported)
    • Cheapest scoring at least as high: Particle-1 Max · Intensive: 71.7 for $0.0256 (provider-reported) (one of 7 such settings)
  2. HighBalanced

    50.6 · rank 44 on the Balanced leaderboard

    Cost per program
    $0.0264 · provider-reported
    Working programs
    8 of 8
    API time
    2.4 min (observed mean, descriptive)

    1.1 points above Low, at 3.1× the cost

    • Higher score for $0.0264 or less: Particle-1 Max · Intensive: 71.7 for $0.0256 (provider-reported)
    • Cheapest scoring at least as high: GLM-5.3 Flash High · Balanced: 60.7 for $0.00579 (provider-reported) (one of 11 such settings)
  3. LowBaseline

    49.5 · rank 29 on the Baseline leaderboard

    Cost per program
    $0.00852 · provider-reported
    Working programs
    8 of 8
    API time
    27 s (observed mean, descriptive)
    • Higher score for $0.00852 or less: GLM-5.3 Flash High · Balanced: 60.7 for $0.00579 (provider-reported)
    • Cheapest scoring at least as high: Muse Ember Medium · Balanced: 49.7 for $0.00550 (provider-reported) (one of 4 such settings)

Score and cost of every ranked setting

The chart needs JavaScript. The setting cards hold this model's results, and Charts compares every model.

Each setting is placed by its published score and its generation cost per player program, the mean over the game cells including repairs and failed attempts. Higher is better; further left costs less (logarithmic axis, with recorded $0 in its own lane). Filled marks are compared; hollow marks are not (the reason is in the table). The line marks the best value across ranked reasoning settings: no ranked setting with complete cost evidence scores at least as high for the same or less. It includes models without an Overall rank, so it can differ from the line on Charts. Costs mix provider-reported charges and list-price estimates; differences in score points are not tested for significance.
Show the numbers

The table needs JavaScript.

Available settings and grouping

Available settings

LowHighMax

Checked on 2026-09-22 · Provider documentation

How settings are grouped

  • Baseline: Low — tested and included
  • Balanced: High — tested and included
  • Intensive: Max — tested and included

Scores by game

  • Go60.1
  • Chess53.6
  • Shogi60.5
  • Xiangqi60.5
  • Hex53.9
  • International Draughts64.4
  • Game of the Amazons64.1
  • Tumbleweed32.7

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: Max (68.6).

Across included settings, the highest game score is International Draughts (64.4); the lowest is Tumbleweed (32.7). 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
z-ai/glm-5.3-flashx
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
131072 (29 of 29 indexed attempts)

Generation evidence

  • Low: 8 contributing player programs across 8 games
  • High: 8 contributing player programs across 8 games
  • Max: 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)
GoIntensive1 working of 1 attempt41.71385.1237 / 3 / 360
GoBalanced1 working of 1 attempt75.51736.6437 / 6 / 157
GoBaseline1 working of 1 attempt63.31597.1306 / 122 / 172
ChessIntensive1 working of 1 attempt67.51609.5237 / 257 / 68
ChessBalanced1 working of 1 attempt52.11479.276 / 406 / 80
ChessBaseline1 working of 1 attempt41.31394.512 / 415 / 135
ShogiIntensive1 working of 1 attempt82.31808.4450 / 7 / 115
ShogiBalanced1 working of 1 attempt53.51446.9252 / 58 / 262
ShogiBaseline1 working of 1 attempt45.61375.7178 / 116 / 278
XiangqiIntensive1 working of 1 attempt76.71709.5343 / 176 / 59
XiangqiBalanced1 working of 1 attempt54.71503.4140 / 324 / 114
XiangqiBaseline1 working of 1 attempt50.21467.885 / 383 / 110
HexIntensive1 working of 1 attempt76.31832.5449 / 0 / 163
HexBalanced1 working of 1 attempt39.41266.9228 / 0 / 384
HexBaseline1 working of 1 attempt46.21355.6268 / 0 / 344
International DraughtsIntensive1 working of 1 attempt79.71953.3442 / 23 / 119
International DraughtsBalanced1 working of 1 attempt53.81483.6283 / 18 / 283
International DraughtsBaseline1 working of 1 attempt59.81585.5327 / 4 / 253
Game of the AmazonsIntensive1 working of 1 attempt70.01697.2410 / 0 / 200
Game of the AmazonsBalanced1 working of 1 attempt58.91533.7337 / 0 / 273
Game of the AmazonsBaseline1 working of 1 attempt63.41596.1366 / 0 / 244
TumbleweedIntensive1 working of 1 attempt54.61464.2316 / 0 / 298
TumbleweedBalanced1 working of 1 attempt17.1886.396 / 0 / 518
TumbleweedBaseline1 working of 1 attempt26.51029.7150 / 0 / 464

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

0 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, an illegal move, or a timeout repeated by a later quiet replay, pinned on one program, also counts as that program's loss in its match record; other faults 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

Max: 100% / 0% / 0% · n 8High: 63% / 37% / 0% · n 8Low: 75% / 25% / 0% · n 8 All three benchmark settings tested

Generation cost, all settings combined

$0.03average estimated cost per game and reasoning setting

Median
$0.03
Range
$0.0054–$0.10
Provider-reported total
$0.76
Cost data
24 combinations · 24 runs
Price source
Provider-reported
Price list
boardgame-list-prices-2026-09-08
Average output
23k tokens
Output range
2.3k–74k tokens
Runs with output-token data
24
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