Atlas

Compare Models

Grok 4.7 vs Hy4

Hy4 leads by 1.3 capability points (≈95% interval −1.2 to +3.8): statistically tied · P(higher) ≈ 84%

Grok 4.7 wins 10 of 16 shared benchmarks (5 lost, 1 tied)

Capability vs. relative cost
Grok 4.7Hy4Closed weightsOpen weights

Hy4 has no setting with a relative cost yet.

Benchmark wins10 – 5 · 16 shared, 1 tied

Comparing matching setups only (162 benchmarks excluded) · suites counted once
← Grok 4.7Hy4 →
Data: Surge AI, Mercor, Collinear AI and 3 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Hy4 minus Grok 4.7.

Chartography (November 2026 rescoring)Perception & media › Documents & charts74.649.2−25.4
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine27.67.2−20.4
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations44.023.7−20.3
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax26.59.2−17.3
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations32.648.1+15.5
CWE-bench v2 Pass@1Software › Cybersecurity54.764.1+9.4
LLM Debate BenchmarkLanguage & writing › Language & wordplay16011647+6.5
RiemannBench (November 2026 rescoring)Math57.651.2−6.4
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following60.866.8+6.0
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal19.513.6−5.9
DAXBench v2Software › Data & ML92.687.2−5.3
GDP.pdf (November 2026 rescoring)Professional › General knowledge work48.843.6−5.2
APEX-Agents 1.1Professional › General knowledge work80.485.1+4.7
GDP.xlsxProfessional › General knowledge work22.317.7−4.6
Design Arena Overall Code CategoriesSoftware › Frontend & design12701260−1.7
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations68.268.20.0
Data: Surge AI, Mercor, Collinear AI and 3 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…