Gemini 4 benchmarked: Neon sweeps Pareto frontiers from $0.29 to $6.44 per task

ArtificialAnlys · x · 2026-11-20

Artificial Analysis compared all 7 Gemini 4 models across Neon and Carbon releases. Gemini 4 Neon (xhigh) tops the Intelligence Index at 61 vs Carbon (high) at 60; Neon and Hydrogen occupy the entire Pareto frontier of intelligence vs cost and vs time per task, scaling from 27 at $0.29/task (Hydrogen, low) to 61 at $6.44/task (Neon, xhigh). Neon (low) runs 3.7 min/task vs Carbon (medium) at 29.4 min. Standout scores: 67 engineering, 71.3% HLE-Diamond, 65.2% Terminal-Bench 4.0. Carbon breaks the scaling trend with stronger results on some hard subsets (69% CritPt v2 hard). Developers can tune effort level for their cost/latency tradeoff.

Related event: Gemini 4 Neon Launches, Topping Coding Agent Index and Reshaping the Cost-Performance Frontier(28 posts)→

Original post →

More from Models

Models channel →