Expired50% confidenceBenchmarkExact time

DeepSeek幻觉相关得分仅约9%，而GLM 5.3 Flash得分高达72%（数字越高越好）

1

Sources

50%

Confidence

Short-term (~14 days)

Relevance

9/11/2026

First Seen

Valid until: 9/25/2026(expired)

Sources

DeepSeek V4 Flash视觉版实测：速度惊艳但设计能力堪忧

Related Entities

DeepSeek V4 Flashmodel GLM 5.3 Flashmodel

Related Claims

UnverifiedDeepSuite基准测试的误报率为44.9%，漏报率为1.2%67% similar Unverified在GPQA V2版本测试中，Glimmer的幻觉率高达82%65% similar Unverified千问3.8 Flash Next的稀疏率约为95%，每次推理只动用不到5%的参数65% similar Verified自回归解码阶段GPU算力利用率往往不足10%，属于内存密集型操作61% similar UnverifiedMuse Glimmer在Artificial Analysis第二版测试中幻觉率高达82%，知识工作和幻觉是其最大弱点61% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/896150

API

curl https://kongchang.com/api/v1/knowledge/claims/896150

MCP

get_claim(id=896150)