Recently, the AI community once again sparked heated discussions over version transparency. The incident began when developer argofowl accidentally discovered while reviewing the official logs of Claude Code that starting from version 2.1.237, the system quietly mapped the user-selected "high" reasoning level directly to the value 10, which was originally associated with the "low" setting, and the official did not mention this change in the update log.
In response to the community's questions and discussions, a Claude Code engineer later stated that this was actually a configuration test for an API service. This value alone does not have practical significance and would not have a substantial impact on overall performance. However, this explanation did not fully ease users' concerns. Subsequently, tech blogger Chubby also publicly pointed out that Opus5 showed a noticeable decline in performance. In response, the relevant engineers admitted that the model currently exhibited some instability and stated that the team had prioritized resolving this issue.
This incident not only exposed the common pain point in the current AI industry where "benchmark scores are severely disconnected from users' actual experiences," but also highlighted the risks of opaque model version updates. As artificial intelligence gradually becomes the infrastructure of the digital world, the continuous stability and transparency of models remain essential cornerstones for maintaining user trust.
