Atlas

Compare Models

Gemini 3.8 Flash vs Hy4

Hy4 leads by 1.2 capability points (≈95% interval −1.3 to +3.7): statistically tied · P(higher) ≈ 83%

Gemini 3.8 Flash wins 10 of 19 shared benchmarks (8 lost, 1 tied)

Capability vs. relative cost
Gemini 3.8 FlashHy4Closed weightsOpen weights

Hy4 has no setting with a relative cost yet.

Benchmark wins10 – 8 · 19 shared, 1 tied

Comparing matching setups only (144 benchmarks excluded) · suites counted once
← Gemini 3.8 FlashHy4 →
Data: Surge AI, Mercor, Neel Nanda and 7 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Hy4 minus Gemini 3.8 Flash.

Chartography (November 2026 rescoring)Perception & media › Documents & charts81.749.2−32.5
Terminal-Bench 4.0Software › Software engineering19.144.4+25.3
LLM Debate BenchmarkLanguage & writing › Language & wordplay14881647+22.7
NoCoT-Bench NCRI 15.2Knowledge & reasoning › Abstract reasoning12694.0−17.4
APEX-Agents 1.1Professional › General knowledge work72.985.1+12.2
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations58.568.2+9.7
Design Arena Overall Code CategoriesSoftware › Frontend & design12991260−6.5
NoCoT-Bench NCKI (kspine v3)Knowledge & reasoning › General knowledge121111−6.4
RiemannBench (November 2026 rescoring)Math56.851.2−5.6
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine2.07.2+5.2
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax4.59.2+4.7
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following63.266.8+3.6
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations26.523.7−2.8
Harvey LAB (Mercor run, undisclosed ~213-task cohort)Professional › Legal11.413.6+2.2
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge34.332.7−1.6
τ³-Banking (Mercor run, bm25_grep retrieval)Tool & computer use › Customer service & operations48.848.1−0.7
GDP.pdf (November 2026 rescoring)Professional › General knowledge work44.243.6−0.6
GDP.xlsxProfessional › General knowledge work18.317.7−0.6
DAXBench v2Software › Data & ML87.287.20.0
Data: Surge AI, Mercor, Neel Nanda and 7 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…