LuminaBench
Mathematics benchmark / Epoch AI

FrontierMath Tier 4 v2

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.

Solve rate

Version FrontierMath-Tier-4-v2-PrivateVerified 6 November 202678 published resultsOfficial source

FrontierMath Tier 4 v2

LuminaBench
Show
20 of 71 models · best variant per model
03006009001,2001,162GPT-6.1 SolMax1,156GPT-6 AstraxHigh1,149Claude Opus 5.5Max1,143GPT-6.1 TerraMax1,137Claude Fable 5Max1,136GPT-6 SolMax1,131Claude Fable 5.1Max1,118GPT-5.6 SolMax1,111Claude Sonnet 5.5Max1,105Gemini 4 ArgonHigh1,105GPT-5.5 ProxHigh1,099gdm-ai-co-mathematician1,092Claude Opus 5Max1,091GPT-5.5xHigh1,086muse-wonder-1.0Max1,086GPT-5.6 TerraMax1,060GPT-5.6 LunaMax1,054GPT-5.4 ProxHigh1,048GPT-6 LunaMax1,048Claude Opus 4.8Max

No comparable per-task resource measurements were published. Missing metrics are omitted from their graph; quality scores remain available.

Results table

Method & coverage

FrontierMath Tier 4 v2 private mathematics problems; Python/tool scaffold. Preserve the v2 cohort, separate from v1 and Tiers 1–3.

No comparable per-task resource measurements were published.

How the Lumina conversion works

1,000 + 75 × ((original score − 37.92010147187062) / 28.654795119029774). 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
frontiermath-tier4-v2-z75-v1 · experimental
Reference mean
37.92010147187062
Population SD
28.654795119029774
Coverage
Standalone benchmark.
Identity, exclusions & attribution

FrontierMath Tier 4 v2 private mathematics problems; Python/tool scaffold. Preserve the v2 cohort, separate from v1 and Tiers 1–3.

Each run ID is retained. Repeated observations are not additional model releases or native reasoning settings. Standard errors are retained only where their mean-score association matches the selected primary metric.

Text mathematical statements/proof/code tools; no image, video, speech or screenshot input protocol.

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

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 source · Publication terms / evidence

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