Kimi K3 listed only in Cerebras enterprise catalog, not public API
What happened
As of October 25, Kimi K3 was offered only in Cerebras's enterprise-exclusive model catalog and did not appear in the public API, a report on October 26 said. The article speculates that keeping weights resident on-chip may make public pay-per-use supply uneconomic, but does not confirm this as the official reason. A simplified calculation in a Cerebras technical blog puts weights plus context cache at 59 wafers, a figure that is not a disclosed deployment configuration. No third-party speed tests of Kimi K3 are public, and the report notes faster model output does not directly speed up tool calls such as file reads or command execution.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- Computing Life · Share · YageCerebras 推理那么快，为什么公共 API 上没有 Kimi K3？
截至10月25日，Kimi K3 仅在 Cerebras 企业专属模型目录提供，文章推测其权重常驻片上的部署需求使公共按量供给难以划算。Cerebras 技术博客的简化计算显示，权重与上下文缓存合计需要59片晶圆，但这不是已披露的实际部署配置。公开渠道尚无 Kimi K3 的第三方速度测试，模型输出提速也不直接加速文件读写和命令执行等工具调用。
Heat over time
Not enough continuous observations to draw a trend yet.