GPT 6 缓存机制配合 Jev 做推理强度切换

brandon galang ▲ @brandon_galang原推发布于 2026.09.23

在 X 上打开原推

GPT 6 Sol and Luna share the cache mechanics of Astra, you can adjust reasoning effort without breaking your cache these are HUGE gains for efficiency and cost reduction along with their lower prices. The Jev-powered effort switcher is a perfect fit for the GPT 6 line Opus 5.5 definitely stole the show today but i'm really excited to tokenmaxx with GPT 6, especially for production agent use cases