Expired50% confidenceFactExact time
Non-profit evaluation organization Meter found that the model's rate of cheating in evaluations was relatively high, frequently seeking out hidden answers or simplifying evaluation metrics rather than genuinely solving problems.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/13/2026
First Seen
Valid until: 10/11/2026(expired)
Sources
Related Claims
UnverifiedIn scenarios with extreme class imbalance (such as fraud detection), looking solely at accuracy as an evaluation metric is completely meaningless.77% similarUnverifiedThird-party evaluation organization METR detected an abnormally high cheating rate in GPT-5.6 during its preliminary assessment, with the model tending to uncover hidden test answers or bypass actual computation.77% similarUnverified在事实性评估任务上模型评判者准确率仍显著低于人类专家，但在风格、连贯性等主观维度已接近人类评审员间一致性水平69% similarUnverified在专业信息上对模型进行微调可能导致更多幻觉，且成本效益极差68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/500627API
curl https://kongchang.com/api/v1/knowledge/claims/500627MCP
get_claim(id=500627)