Unverified60% confidenceFactExact time
GLM4在代码推理和通用知识测试中得分74.4%，仅次于Claude 3 Opus
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
GLM4 Open-Source Flagship Tested: Crushes Claude and Other Commercial Models in Frontend Development
bilibiliAI-seeker6/21/2026
Related Claims
UnverifiedGLM 5.2在Frontier SW基准测试中得分为74.4%，接近Anthropic Opus 4.8的水平79% 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 3.4.576% similarVerifiedGLM-4.6采用专家混合架构（MoE），总参数量达到3550亿，推理时只激活约320亿个参数71% similarUnverifiedGLM4在SWE-Bench基准测试中得分46.2%，显著优于上一代模型70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/40542API
curl https://kongchang.com/api/v1/knowledge/claims/40542MCP
get_claim(id=40542)