Unverified50% confidenceFactExact time
GPT-4o mini、Claude Haiku等轻量模型采用蒸馏技术，在参数量缩减80%以上情况下仍能保留旗舰模型60-70%的综合能力
1
Sources
50%
Confidence
Long-term
Relevance
7/12/2026
First Seen
Sources
实测GPT-5.6三模型Sol/Terra/Luna：编程能力几何？
bilibili神烦老狗7/10/2026
Related Claims
Unverified轻量模型如GPT-4o mini、Claude Haiku保持低价或订阅覆盖，旗舰模型则逐步向按量付费或高端订阅层迁移80% similarUnverifiedClaude Haiku 4.5相比GPT-4o mini等同级别模型在定价上仍有明显优势79% similarUnverified旗舰模型（如GPT-4o、Claude 3.5 Sonnet）与轻量模型（如GPT-4o-mini、Claude Haiku）在价格上可相差10-20倍，但在大量常规任务上性能差距并不显著78% similarVerifiedClaude Haiku和GPT-4o mini被归类为轻量级模型，推理速度快、API调用成本低，适合处理格式转换和简单重构等任务76% similarUnverifiedGPT-5 Mini在同类测试中表现远优于Claude 4.5 Haiku75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/492589API
curl https://kongchang.com/api/v1/knowledge/claims/492589MCP
get_claim(id=492589)