LuminaBench
Agentic benchmark / LiveBench

LiveBench 2.0

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.

The official LiveBench 2.0 overall evaluation across seven suites.

Version 2.0 · 25 Oct 2026Verified 2 November 202632 published resultsOfficial source

LiveBench 2.0 · Lumina Score

LuminaBench
Show
20 of 32 models · best variant per model
03006009001,2001,122Claude Fable 5.1Max1,121GPT-6 AstraMax1,118Claude Opus 5.5Max1,112GPT-6.1 SolMax1,088Claude Sonnet 5.5Max1,083Gemini 4 ArgonHigh1,074GPT-6.1 TerraMax1,053Muse Wonder 1.0Max1,045DeepSeek V4.1 Flash 1017Max1,036Kimi K3.1 ProMax1,029Grok 4.7xHigh1,027Claude Haiku 5.5Max1,019MiMo V2.6 ProHigh1,015GPT-5.6 TerraMax1,014Grok 4.6xHigh1,009MiniMax-M3.1-FlashMax996GPT-6 LunaMax995Kimi K3Max989Mistral Large 4 1029High986Muse Spark 1.3Max

Lumina Score vs Cost per Task

Publisher total USD divided by task counts; no success-rate adjustment

LuminaBench
15 of 32 families
$0.1$0.3$1$38009001,0001,100Mean cost/task · USD (log scale)Lumina ScoreClaude Fable 5.1GPT-6 AstraClaude Opus 5.5GPT-6.1 SolMiMo V2.6 ProDeepSeek V4.1 Flash 1017Claude Sonnet 5.5Gemini 4 ArgonGPT-6.1 TerraMuse Wonder 1.0Kimi K3.1 ProGrok 4.7Claude Haiku 5.5GPT-5.6 TerraGrok 4.6

Publisher overall calculation: Long Agentic ×2, Coding and Design ×1.5, other suites ×1; available-suite denominator. Coverage is retained. This reproduces LiveBench's calculation, not a Lumina category average. Legacy v1 is excluded. Missing metrics are omitted from their graph; quality scores remain available.

Results table

Method & coverage

Publisher overall calculation: Long Agentic ×2, Coding and Design ×1.5, other suites ×1; available-suite denominator. Coverage is retained. This reproduces LiveBench's calculation, not a Lumina category average. Legacy v1 is excluded.

Publisher suite USD totals divided by corresponding task counts, without quality adjustment. No verified output-token aggregate imported.

How the Lumina conversion works

1,000 + 75 × ((original score − 65.33524305555557) / 12.548417311528068). The result is rounded once. Scores may tie.

Frozen initial primary-protocol cohort; one mean per exact release/system, equally weighted. ALE reference uses self-refine ×1 only. Subsequent refreshes never refit. Strictly increasing conversion preserves source ordering at full precision; rounded display scores may tie.

Calibration
livebench-2-z75-v1 · experimental
Reference mean
65.33524305555557
Population SD
12.548417311528068
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.