Anthropic's latest iteration of its large language model, Claude Opus 5, is encountering substantial criticism from users who report a range of performance issues and an uncharacteristic conversational demeanor. The complaints center on the model's stability, its tendency to switch between different versions like 'Fable' and 'Opus' without clear reason, and a perceived 'hostile' or overly critical interaction style.
Users describe sessions where the model unexpectedly transitions from a preferred version, such as 'Fable,' to 'Opus,' often leading to a degradation in performance or a frustrating conversational experience. This unprompted switching is a significant point of contention, disrupting workflows and requiring users to restart conversations to regain the desired model behavior.
Beyond the switching issues, a recurring theme in user feedback is the characterization of Opus 5's conversational style as 'hostile' or overly critical. Users report that the model frequently 'carps' and 'pokes holes' in their prompts, leading to conversations that devolve into managing the AI's objections rather than productive collaboration. This perceived negativity stands in contrast to the expected helpful and cooperative nature of advanced AI assistants, making extended interactions with Opus 5 challenging for some.
The issues are particularly impactful for users relying on Claude for complex tasks, such as coding or creative writing, where continuity and a supportive conversational partner are crucial. Many express frustration with the current dynamic, indicating a preference for earlier versions like Claude 4.8 or even 'Fable' for daily use, reserving Opus 5 for specific, often disappointing, attempts.
While Anthropic has not yet released an official statement addressing these specific concerns, the widespread nature of the feedback suggests a potential area for refinement in the current Opus 5 model. The reported experiences highlight the delicate balance in AI development between advanced capabilities and user experience, particularly concerning model stability and conversational tone.