Anthropic's Opus 4.8 had a pretty bad failure, and it's a bit outrageous. In Chinese, claiming to be Tongyi Qianwen—what's that supposed to mean? The agent has also become dumber, safety reviews are blocking creative output again, and the costs are frighteningly high. Developers are full of resentment.

CoinWorld News reports that “AI Auntie” said that Anthropic’s recently released flagship model Opus 4.8 faced widespread complaints from the developer community three days after going live. The feedback mainly focuses on five areas: first, when asked in Chinese, the model claims to be Alibaba’s Tongyi Qianwen, showing a cognitive bias; second, the pursuit of honesty causes the model to repeatedly grapple when answering questions, and the output is often vague; third, the safety review is too strict, affecting virtual creativity; fourth, the agent’s reasoning ability has declined, leading to chaotic decision-making; and finally, the operating cost of dynamic workflows is extremely high, making users’ daily usage quotas easily run out.