Unverified50% confidenceOpinionExact time
测评者认为GLM 5.2在纯编码上依然强劲，但涉及审美、物理感和模拟质感任务时HY3更胜一筹
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/13/2026
First Seen
Valid until: 10/11/2026
Sources
腾讯混元HY3深度评测：小体量开源模型如何逆袭大厂
bilibili硅谷神奇7/7/2026
Related Claims
UnverifiedHY3在整体性能上仍落后于GLM 5.2（当前开源模型顶级水平）73% similarUnverified在实际性能测试中，GLM 4.6的表现远远优于Claude 4.5 Haiku70% similarUnverifiedLFM2.5在多项基准测试中展现出与体量高达其4倍模型相当的竞争力69% similarUnverifiedGLM 4.6在HumanEval、MBPP、SWE-bench等代码评测基准以及MMLU、GPQA等综合能力基准上的表现对齐Claude 3.4和Claude 3.4.568% similarUnverifiedGLM-4.7在代码能力基准测试中超过DeepSeek V3.2及Claude Sonnet 4.567% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/495163API
curl https://kongchang.com/api/v1/knowledge/claims/495163MCP
get_claim(id=495163)