GPT-5.6 Luna Max in Codex Outperforms Sol at Fraction of the Cost

DeryaTR_ · x · 2026-08-03

A user reports remarkable results using the unpublicized GPT-5.6 Luna Max model within Codex. With detailed prompts, its output quality sometimes surpasses Sol light or Sol medium, while costing 5-6x less in tokens. On a Pro account, users can run multiple continuous threads without hitting limits, though Sol models remain better suited for planning and auditing tasks.

Original post →

More from coding & agent

coding & agent channel →