Testing GPT image tiers: high vs ultra barely differ, quality drops hard on low settings

kevinkern · x · 2026-09-13

Browsing the GPT Observ site, the author compared image quality tiers: high and ultra look nearly identical, while medium and especially low show visible quality drops. The sweet spot appears to be around Med/High. Useful guidance for users paying per-generation.

Original post →

More from Multimodal

Multimodal channel →