Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
@teortaxesTex
> the flagship model pushes hardcore capabilities, while lightweight K2.8 handles programming …
given that even the "lightweight" K2.8 is (if it's another K2+) way more expensive at inference than *ANY* DeepSeek since April, I'm worried for Moonshot.
Might have scaled too early.
given that even the "lightweight" K2.8 is (if it's another K2+) way more expensive at inference than *ANY* DeepSeek since April, I'm worried for Moonshot.
Might have scaled too early.
xiax｜AI & Agents@xiax0603 · Oct 10🚨重磅爆料｜国内AI赛道又迎来关键节点！月之暗面的Kimi K3.1已经敲定10月中上旬上线。重点打磨长程Agent能力。
月之暗面两套模型同步推进，不再死磕加大参数量。他们用电脑操作记录+真实任务数据做训练，专门解决长流程任务容易断档的痛点。
月之暗面两套模型同步推进，不再死磕加大参数量。他们用电脑操作记录+真实任务数据做训练，专门解决长流程任务容易断档的痛点。
Open quoted post →
14 172