模型級聯 / 降級鏈

ML基礎設施

查詢先送給快速且經濟的小模型，若信心度不足則自動升級到大模型，兼顧成本效率和回應品質。

agentsystem

為什麼需要 OSOP

模型級聯是成本優化的關鍵策略。OSOP 用條件邏輯和降級邊定義模型選擇流程，讓你輕鬆調整閾值和模型組合，同時記錄每次路由決策以利分析優化。

Receive Query

event

Fast Model (Haiku)

agent

Confidence Check

system

Large Model (Opus)

agent

Return Response

api

Receive Query→Fast Model (Haiku)sequential

Fast Model (Haiku)→Confidence Checksequential

Confidence Check→Return Responseconditionalconfidence >= 0.8

Confidence Check→Large Model (Opus)conditionalconfidence < 0.8

Large Model (Opus)→Return Responsesequential

Steps

Connections

Node Types