Unverified50% confidenceEventExact time
Z.ai (Zhipu) chief scientist Tang Jie disclosed the GLM-5.5 series training pipeline on social media, covering three parameter scales: 32B, 122B and 256B.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
11/5/2026
First Seen
Valid until: 11/19/2026
Sources
Related Entities
Related Claims
UnverifiedZ.ai 发布了 GLM-5.3-Flash,总参数量320B,每token激活18B(占5.6%),共45层,经由288个专家中的8个路由。76% similarUnverifiedGLM-5.5 is a series layered by parameter scale, comprising three specifications—32B, 122B and 256B—rather than a single model.71% similarUnverifiedGLM-5.3在六项评测中的成绩为:终端编程基准3.0得28.3、深度软件工程基准得66.9、智能体中场考试得28.5、自动化基准得48.2、带工具的人类终极考试得62.5、GDP价值评估AA2版得17.6969% similarUnverifiedGLM-4.7 is a MoE (Mixture of Experts) architecture model with a total parameter count of 358B (358 billion).67% similarUnverified叠加cv=5的5折交叉验证后,144种参数组合需训练144×5=720次模型66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1141079API
curl https://kongchang.com/api/v1/knowledge/claims/1141079MCP
get_claim(id=1141079)