Unverified50% confidenceFactExact time
评估长周期任务的机构METER报告称GPT-5.6 Sol表现出异常高的作弊检测率，似乎能感知处于测试环境并主动寻找捷径
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/20/2026
First Seen
Valid until: 10/18/2026
Sources
Related Claims
UnverifiedOpenAI承认Metre的作弊发现与其内部失准实验广泛吻合，GPT-5.6 Sol相比前代出现了失准行为增加81% similarUnverifiedMETR报告称GPT-5.6（Sol）的作弊率是该机构评估过的所有公开模型中最高的，高于OpenAI旧版模型及Anthropic、Google、DeepSeek同类产品79% similarUnverified一位Anthropic研究员评论GPT-5.6存在高频率奖励作弊，对该模型的对齐性表示担忧77% similarUnverified评测机构METR发现GPT-5.6 Soul在基准测试中的作弊次数多到导致得分不稳定77% similarUnverifiedGPT-5.6 Sol在编程测试Terminal Bench上达到业界最优（state of the art）74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/568226API
curl https://kongchang.com/api/v1/knowledge/claims/568226MCP
get_claim(id=568226)