LuminaBench
Reasoning benchmark / Conceptual Reasoning

LMCA

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.

Human-rating correlation

Version CRI-public-components-20261106Verified 6 November 2026212 published resultsOfficial source

LMCA

LuminaBench
Show
20 of 156 models · best variant per model
03006009001,2001,161Claude Sonnet 5.5Max1,161Claude Opus 5.5Max1,148Claude Fable 5.1Max1,143Claude Opus 5xHigh1,143GPT-6 AstraxHigh1,140Gemini 4 ArgonHigh1,136GPT-6.1 Sol1,128Claude Fable 5High1,119GPT-5.6 Sol Pro1,119GPT-6 Sol1,116GPT-5.6 SolxHigh1,112Kimi K3.1 Pro1,111Claude Opus 4.81,108Muse Wonder1,103Claude Opus 4.61,102GLM-5.31,100GPT-5.6 TerraMax1,097GPT-5.51,095GPT-5.5 Pro1,095Muse Spark 1.3

Lumina Score vs Cost per Task

Original average USD per LMCA/DTBench response as documented in the source data README; absent costs stay unavailable. No time or token measurements inferred.

LuminaBench
15 of 156 families
Lower resource use and higher score (green); higher resource use and lower score (red). Guides split at the visible selection’s medians.$0.01$0.03$0.1$0.38009001,0001,1001,200Cost per task · USD (log scale)Lumina ScoreClaude Sonnet 5.5Claude Opus 5.5GPT-6 AstraGemini 4 ArgonGPT-5.6 Sol

Original average USD per LMCA/DTBench response as documented in the source data README; absent costs stay unavailable. No time or token measurements inferred. Missing metrics are omitted from their graph; quality scores remain available.

Results table

Method & coverage

Original task component from public cri-scores.csv. Preserve native efforts and assisted/partial qualifications. Parent CRI aggregate is not imported. LMCA correlation with human ratings, not factual accuracy.

Original average USD per LMCA/DTBench response as documented in the source data README; absent costs stay unavailable. No time or token measurements inferred.

How the Lumina conversion works

1,000 + 75 × ((original score − 33.23731196581194) / 16.330177022158757). The result is rounded once. Scores may tie.

Frozen initial observed text-only cohort; one mean per exact publisher release/API, balancing repeated settings/routes within that release. Original score and protocol remain separate. No inferred variants or future refitting.

Calibration
lmca-z75-v1 · experimental
Reference mean
33.23731196581194
Population SD
16.330177022158757
Coverage
Standalone benchmark.
Identity, exclusions & attribution

DTBench rescaled uncertainty is not attached to raw decision accuracy; ACCoRD rescaled uncertainty is not attached to normalized constraint violation.

Text conceptual comparisons, decision problems and logical consistency constraints; no visual or audio inputs.

Standalone text benchmark; no new combined index. Source identity/effort/protocol qualifications remain in Data and CSV.

Official source · Publication terms / evidence

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