Back to explore

Using Jev to Decide Model Switching in Codex Tasks Cuts Astra Cost by 50%

A report claims that using Jev to decide which model to use for current inference during Codex tasks cut Astra cost by 50%. The comments are debating whether it breaks caching, and the poster plans to test and share results when time permits.

「Codexのタスク中にどのモデルを今の推論に使うべきかの切替判定にJevを使って、Astraコストが50%低下。」と言う報告。 キャッシュ壊す壊さないでコメント欄が炎上中だけど、実際にやってみれば良いよね。私も時間出来たらテストして共有します☺️

· 0 likesOpen on X