LuminaBench
Reasoning benchmark / Arena

Arena Text — Style Controlled

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.

Overall text preferences, controlled for response style.

Version Overall · Style controlledVerified 2 November 2026419 published resultsOfficial source

Arena Text — Style Controlled · Lumina Score

Style-controlled overall · Published 2026-10-28

LuminaBench
Show
20 of 404 models · best variant per model
03006009001,2001,120Gemini 4 CarbonHigh1,116Gemini 4 ArgonHigh1,108Claude Opus 5.5High1,107Muse WonderMax1,106Claude Fable 5.1Max1,105Claude Fable 5High1,105Claude Opus 4.6High1,104Claude Opus 4.7High1,101Gemini 3.8 FlashHigh1,100Muse Spark 1.3Max1,098Muse Spark 1.2xHigh1,098Claude Opus 5High1,097Muse Spark 1.11,097Gemini 3.7 FlashHigh1,096Muse Spark1,096Kimi K3Max1,095Gemini 3.1 Pro Preview1,094Gemini 3 Pro1,094GPT-5.6 SolxHigh1,093Gemini 3.6 FlashHigh

Publisher preference ratings, bounds and vote counts retained. Missing metrics are omitted from their graph; quality scores remain available.

Results table

Source & conversion

Arena text_style_control, overall only. Published 2026-10-28; underlying evaluation timestamps are not supplied.

Preference ratings with exact publisher bounds, variance and vote counts. The dataset does not specify the interval level. No resource telemetry is published.

How the Lumina conversion works

1,000 + 75 × ((original score − 1334.3798242242701) / 120.5505947005131). The result is rounded once. Scores may tie.

Frozen initial valid overall cohort; one arithmetic mean per exact evaluated release/system, equal weight across releases rather than effort rows. Unbounded rating/IPS scales use a broad finite numeric guard, not percentage bounds or sign clamping. Subsequent refreshes never refit this calibration. Conversion is not an estimate or combined index.

Calibration
arena-text-style-control-z75-v1 · experimental
Reference mean
1334.3798242242701
Population SD
120.5505947005131
Coverage
Standalone benchmark.
Identity, exclusions & attribution

Exact publisher labels and release qualifiers are retained. Only explicit native effort tokens create reasoning variants; token budgets remain separate.

Agent orchestrator, routed, dated and harness-qualified systems are not automatically equated with naked base models.

0 ambiguous repeated records are retained in original.json but excluded from ranking.

Official dataset · Pinned source revision · CC BY 4.0

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