Unverified70% confidenceFactTime unknown

GLM5的上下文长度超过20万token时会出现明显的上下文腐烂问题

1

Sources

70%

Confidence

Medium-term (~90 days)

Relevance

6/1/2026

First Seen

Valid until: 8/30/2026

Sources

GLM5编程方案实测：年费仅$84，配置教程与使用技巧全攻略

bilibiliAI-seeker

Related Entities

Related Claims

UnverifiedGLM5.2的推理速度明显慢于同级别模型，是其最大短板68% similar UnverifiedGLM-5.2 在 Deeps-WE 基准上从上一代的18提升到4667% similar UnverifiedGLM 4.6相较于GLM 4.5，代码能力提升27%，token消耗降低30%65% similar UnverifiedGLM-4.6的激活比例约为9%，在当前MoE模型中属于较为激进的稀疏设计64% similar UnverifiedGLM-5.2 在 Frontiers-WE 基准上从上一代的30出头提升到7463% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/19317

API

curl https://kongchang.com/api/v1/knowledge/claims/19317

MCP

get_claim(id=19317)