Opus 5.5 medium beats GPT-6 Astra max 54.5% vs 53.3% at one-fifth the cost per task

haider1 · x · 2026-09-23

Citing a benchmark, haider1 does the math: Opus 5.5 medium scores 54.5% at $0.8/task, while GPT-6 Astra max scores 53.3% at $4.3/task — Anthropic's mid-tier model wins on accuracy while being roughly 5x cheaper per task, a stark value-gap data point in the frontier pricing race.

Original post →

More from Models

Models channel →