LuminaBench
Reasoning benchmark / Arena

Arena Search

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.

Style-controlled preferences for search and grounding systems.

Version Overall · Style controlledVerified 2 November 202642 published resultsOfficial source

Arena Search · Lumina Score

Style-controlled overall · Source date tag 2026-10-13

LuminaBench
Show
20 of 42 models · best variant per model
03006009001,2001,127GPT-5.6 Sol · SearchxHigh1,112GPT-6.1 Sol · SearchMax1,106GPT-6 Astra · SearchMax1,082Claude Fable 5.1 · SearchHigh1,075qwen3.8 max · Search1,074Claude Fable 5 · SearchHigh1,072GPT-6 Sol · SearchMax1,069GPT 5.5 search1,061Claude Opus 4 6 search1,047Claude Opus 5 · SearchHigh1,041Claude Opus 5.5 · SearchMedium1,038Claude Opus 4 7 · Search1,037GPT-6 Luna · SearchMax1,032Gemini 3.1 pro grounding1,032Grok 4.20 Multi-Agent Beta (0309)1,032Grok 4.5 · Search1,024Claude Sonnet 4 6 search1,021Gemini 3 pro grounding1,019grok 4.20 beta1 · Search1,013GPT 5.4 search

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

Results table

Method & coverage

Arena search_style_control, style-controlled overall only. Publisher date tag 2026-10-13; individual run dates and exact tool builds are not published. Search/grounding/document setup remains part of the evaluated system. Separate boards are never combined.

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 − 1189.1332437621043) / 40.573202925110685). 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-search-style-control-z75-v1 · experimental
Reference mean
1189.1332437621043
Population SD
40.573202925110685
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.