Atlas

Compare Models

Hy4 vs Muse Wonder

Muse Wonder leads by 1.6 capability points (≈95% interval −1.0 to +4.1): statistically tied · P(higher) ≈ 89%

Wins 12 of 18 shared benchmarks (6 lost)

Capability vs. relative cost
Hy4Muse WonderClosed weightsOpen weights

Hy4 has no setting with a relative cost yet.

Benchmark wins6 – 12 · 18 shared

Comparing matching setups only (159 benchmarks excluded) · suites counted once
← Hy4Muse Wonder →
Data: Surge AI, Mercor, Neel Nanda and 6 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Muse Wonder minus Hy4.

DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax9.238.8+29.6
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine7.231.2+24.0
GDP.pdf (November 2026 rescoring)Professional › General knowledge work43.654.8+11.2
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal13.623.0+9.4
APEX-Agents 1.1Professional › General knowledge work85.177.0−8.1
RiemannBench (November 2026 rescoring)Math51.259.2+8.0
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations68.261.5−6.7
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations23.729.5+5.8
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge32.737.3+4.6
GDP.xlsxProfessional › General knowledge work17.721.8+4.1
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations48.151.9+3.8
DAXBench v2Software › Data & ML87.284.0−3.2
Agents' Last Exam v1.1 (ALE v1.1)Professional › General knowledge work41.038.5−2.5
NoCoT-Bench NCRI 15.2Knowledge & reasoning › Abstract reasoning94.097.7+2.0
CWE-bench v2 Pass@1Software › Cybersecurity64.166.1+2.0
NoCoT-Bench NCKI (kspine v3)Knowledge & reasoning › General knowledge111109−1.3
LLM Debate BenchmarkLanguage & writing › Language & wordplay16471637−1.3
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following66.867.9+1.1
Data: Surge AI, Mercor, Neel Nanda and 6 othersatlas.village.dev
Win counts use each benchmark’s declared score direction. Represented suite slices and the derived Capability score are excluded from headline totals. Missing or incompatible setups are excluded, never counted as losses.
Loading Atlas data…