Unverified50% confidenceFactExact time
早在2023年，Martian、RouteLLM等项目就在探索用小模型替代大模型完成路由决策以降低推理成本
1
Sources
50%
Confidence
Long-term
Relevance
7/9/2026
First Seen
Sources
Sakana Fugu vs GLM 5.2 Hands-On Test: Who Can Truly Challenge Top-Tier AI Models?
bilibili赛博门外憨7/7/2026
Related Claims
Verified模型蒸馏是将大模型的能力压缩到小模型中以降低推理成本的技术69% similarUnverified2024年以来，Mistral、Phi、Gemma等小型模型通过更精细的数据筛选、知识蒸馏和训练策略优化，在特定任务上逼近甚至超越了体量数倍于己的大模型66% similarUnverified新推出的 Ultra 模式可让模型获得远超常规限制的推理预算，适合处理高难度项目65% similarUnverified候选人曾用一个提升模型加概率校准替换由三个深度模型组成的集成方案，效果同样出色且降低了维护成本65% similarVerifiedThe Cascading model strategy first has a small model handle requests, and when confidence falls below a threshold, it automatically escalates to a large model for reprocessing, in order to maximize cost efficiency.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/367451API
curl https://kongchang.com/api/v1/knowledge/claims/367451MCP
get_claim(id=367451)