Unverified50% confidenceBenchmarkExact time
In DoGBench testing, no single model was able to score above 50%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/2/2026
First Seen
Valid until: 12/31/2026
Sources
Related Entities
Related Claims
UnverifiedItaSoRL实验中智能体的内部状态探测准确率仅约50%，等同于随机猜测水平66% similarUnverifiedAnthropic的研究显示，即使面对毫无根据的质疑，在某些测试中模型被用户简单反驳后放弃正确答案的比例超过50%65% similarUnverifiedThe gap between selection and generation exceeds 50 percentage points, revealing that reliability is far lower than benchmarks suggest when users need the model to directly write out kinship terms.65% similarUnverifiedResearch shows that when the false positive rate exceeds 50%, the practical value of quality tools approaches zero.64% similarUnverifiedOSWorld 2.0中模型评估仅占总得分的11.53%，且没有任何任务的模型评估占比超过50%63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/966331API
curl https://kongchang.com/api/v1/knowledge/claims/966331MCP
get_claim(id=966331)