开发者 Shane Melaugh 在社交媒体上声称,Anthropic 的 Claude Opus 5.5 模型在2024年11月3日后疑似被“削弱”(nerfed)1。尽管模型名称未变,但其推理方式发生了显著变化 1。Melaugh 指出,该声明并非基于专业 AI 评测员或科学基准测试,而是源于个人使用体验落差 1。
他通过让 Gemini 4 Argon 分析自己的 1,400 个 Claude Code 会话发现,高算力模式(high)反应过快导致错误,而超高算力模式(xhigh)仍能保持高效解决问题 1。具体表现为,“high”模式在处理真实 Bug 时耗时3小时无果,而“xhigh”模式几分钟内找到修复方案 1。
Developer Shane Melaugh has alleged that Anthropic’s Claude Opus 5.5 model was potentially "nerfed" following a shift in behavior observed after November 3 1. He stated that although the model name remained unchanged, its reasoning patterns underwent a significant transformation 1. Melaugh clarified that his claims are based on personal user experience rather than professional AI evaluation or scientific benchmarking 1.
To investigate these changes, Melaugh used Gemini 4 Argon to analyze approximately 1,400 of his own Claude Code sessions 1. The analysis revealed distinct performance differences between operational modes: the high-compute mode exhibited overly rapid responses that resulted in errors, whereas the xhigh-compute mode continued to solve problems efficiently 1. Specifically, when addressing real bugs, the high mode reportedly spent three hours without reaching a resolution, while the xhigh mode identified a fix within minutes 1.
评论
还没有评论,欢迎留下第一条。