Unverified50% confidenceFactExact time

DeepSeek等一些开源或半开源模型选择直接展示完整推理过程

1

Sources

50%

Confidence

Medium-term (~90 days)

Relevance

8/13/2026

First Seen

Valid until: 11/11/2026

Sources

deep_think工具意外泄露大模型隐藏推理链：思维链安全边界的脆弱性

Related Claims

UnverifiedDeepSeek V3.2是开源社区首次在推理能力上真正逼近闭源模型天花板的模型80% similar Partially Verified梁文锋表示DeepSeek最强的模型也会开源，且开源模型与自部署模型完全一致79% similar VerifiedDeepSeek 选择开放模型权重79% similar UnverifiedDeepSeek主打推理和编码能力，通义千问押注开源生态77% similar UnverifiedDeepSeek凭借MoE架构的高效推理，通义千问凭借完整的模型尺寸矩阵和宽松的商用协议构成国产开源阵营77% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/741020

API

curl https://kongchang.com/api/v1/knowledge/claims/741020

MCP

get_claim(id=741020)