Expired50% confidenceBenchmarkExact time
In Theo's history audit, Opus 5 aggressively killed the wrong processes, killing more processes in two days than he had with all other tools combined.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
8/25/2026
First Seen
Valid until: 9/8/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedOpus performed the worst in terms of tool misuse, while Opus 5 and 4.8 were the most aggressive in over-building.61% similarUnverified研究表明即便是顶级模型，在工具调用链超过5步后幻觉率会显著上升，原因是早期错误随上下文传递被后续步骤合理化57% similarUnverifiedAnthropic's Opus 4.7 was reportedly subject to compute restrictions and compressed thinking time, prompting large numbers of users to roll back directly to version 4.656% similarUnverifiedThe hallucinated-field problem is more severe on newer Anthropic models (Opus 4.8 and Sonnet 5) than on older models.55% similarUnverifiedOpus 5擅长发现漏洞但利用漏洞的能力被大幅削弱，属于内化式对齐55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/797938API
curl https://kongchang.com/api/v1/knowledge/claims/797938MCP
get_claim(id=797938)