LuminaBench
Coding benchmark / DuelLab

GameBench 2

Lumina VerifiedLumina checked the published sources, versions, model labels, score extraction and conversion. External evaluations were run by their publishers; Lumina has not independently rerun or reproduced them.

AI-written game players evaluated through deterministic matches.

Version 2 · 8 games · B1 v19Verified 2 November 2026231 published resultsOfficial source

GameBench 2 · Lumina Score

LuminaBench
Show
20 of 89 models · best variant per model
03006009001,2001,194Claude Opus 5.5xHigh1,192Claude Fable 5.1xHigh1,192GPT-6 AstraxHigh1,183GPT-6.1 SolxHigh1,181Claude Sonnet 5.5xHigh1,175Pareto 26.10Provider default1,173Gemini 4 ArgonHigh1,171Pareto 26.10 PreviewProvider default1,156Claude Opus 5xHigh1,155Kimi K3.1 ProMax1,137GPT-5.6 SolxHigh1,133GPT-6 SolxHigh1,126Muse WonderxHigh1,118Claude Fable 5xHigh1,104GLM-5.3Max1,104Grok 4.7Medium1,103MiMo-V2.6-ProReasoning enabled1,097DeepSeek V4.1 Flash 1017Max1,094Grok 4.6Medium1,094Muse Spark 1.3xHigh

Lumina Score vs Cost per Player

Publisher estimates with complete cost coverage; pricing bases vary

LuminaBench
15 of 63 families
$0$1$2$3$4$58009001,0001,1001,200Cost/player · USDLumina ScorePareto 26.10Pareto 26.10 PreviewGLM-5.3GPT-6 SolClaude Opus 5.5Claude Fable 5.1GPT-6 AstraGPT-5.6 SolClaude Fable 5Grok 4.7

Publisher-ranked Reasoning Variants only: mean headline score across eight games, including model-caused generation failures. Working-program strength and best-game scores stay separate. Provisional rows remain in source evidence without a Lumina rank. No model-level confidence interval is published. Missing metrics are omitted from their graph; quality scores remain available.

Results table

Method & coverage

Publisher-ranked Reasoning Variants only: mean headline score across eight games, including model-caused generation failures. Working-program strength and best-game scores stay separate. Provisional rows remain in source evidence without a Lumina rank. No model-level confidence interval is published.

Estimated USD/player only where publisher cost coverage is complete. Mixed reconciled/provider/list-price bases retained; incomplete-cell means remain raw supporting evidence. No measured tokens or time imported.

How the Lumina conversion works

1,000 + 75 × ((original score − 48.355259457093695) / 18.13222007063174). The result is rounded once. Scores may tie.

Frozen initial eligible publisher cohort; one arithmetic mean per exact release/system, equally weighted across releases. Subsequent refreshes never refit. Strictly increasing conversion preserves source ordering at full precision; rounded display scores may tie.

Calibration
gamebench-2-z75-v1 · experimental
Reference mean
48.355259457093695
Population SD
18.13222007063174
Coverage
Standalone benchmark.
Identity, exclusions & attribution

Exact source release, effort, route and harness retained. Unknown evaluation dates stay unknown; no cross-source score estimates or combined indexes.

Code licensing is not assumed to clear separately published result redistribution.

All published reasoning settings, including Claude Max, are included by default. Display labels use Low and xHigh; source labels remain in the retained records.

Official results

Publisher terms. Scores and resource figures are attributed to the original publisher.