Atlas

Compare Models

GPT-6.2 Astra vs Hy4

GPT-6.2 Astra leads by 6.7 capability points (≈95% interval +3.8 to +9.5): a clear lead over Hy4 · P(higher) ≈ 100%

Wins 11 of 12 shared benchmarks (1 lost)

Capability vs. relative cost
GPT-6.2 AstraHy4Closed weightsOpen weights

Hy4 has no setting with a relative cost yet.

Benchmark wins11 – 1 · 12 shared

Comparing matching setups only (182 benchmarks excluded) · suites counted once
← GPT-6.2 AstraHy4 →
Data: Surge AI, Lech Mazur Benchmarks, Artificial Analysis and 3 othersatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Hy4 minus GPT-6.2 Astra.

Chartography (November 2026 rescoring)Perception & media › Documents & charts90.749.2−41.5
DAYJOB: Finance (October 2026 rescoring)Professional › Finance, accounting & tax45.89.2−36.6
Humanity's Last Exam DiamondKnowledge & reasoning › General knowledge66.932.7−34.2
DAYJOB: Healthcare (October 2026 rescoring)Health & medicine34.47.2−27.2
RiemannBench (November 2026 rescoring)Math76.051.2−24.8
EnterpriseBench: CoreCraftTool & computer use › Customer service & operations43.668.2+24.6
HANDBOOK.md Agents v1.1Tool & computer use › Customer service & operations45.823.7−22.1
GDP.xlsxProfessional › General knowledge work29.717.7−12.0
Agents' Last Exam v1.1 ScoreProfessional › General knowledge work78.370.3−8.0
GDP.pdf (November 2026 rescoring)Professional › General knowledge work49.243.6−5.6
LLM Debate BenchmarkLanguage & writing › Language & wordplay16731647−3.8
ComplexConstraints (November 2026 rescoring)Language & writing › Instruction following69.666.8−2.8
Data: Surge AI, Lech Mazur Benchmarks, Artificial Analysis 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…