GPT-6 Sol and Luna halve token pricing, pushing OpenAI's cost-efficiency frontier

ArtificialAnlys · x · 2026-09-23

Artificial Analysis reports that OpenAI's new GPT-6 Sol and Luna models match their predecessors' Intelligence Index scores while cutting token pricing by 50%, halving cost per intelligence-index task and pushing OpenAI's cost-efficiency frontier.

The same update rolled out Intelligence Index v4.3 — swapping τ³-Banking for AutomationBench-AA and upgrading Terminal-Bench to 4.0 — where Claude Opus 5.5 takes the top spot. Evaluations also cover GPT-6 Sol/Luna across reasoning-effort tiers (low to xhigh, plus non-reasoning) and Claude Opus 5.5's Adaptive Reasoning configurations.

Original post →

More from Models

Models channel →