Kimi's K2.8 rumored to be post-trained K2 as Moonshot struggles with new base model

Sources suggest Kimi's K2.8 is merely a post-trained version of the older K2 base, indicating Moonshot has struggled to pretrain a new foundation model, while DeepSeek's compute autonomy is seen as a key advantage.

2026-09-11 ~ 2026-09-11 · 2 related posts

• Teortaxes: Kimi stuck serving old K2 base as K2.8 rumored post-trained variant; DeepSeek's compute moat — teortaxesTex · 2026-09-11
• Kimi's base-model struggles vs DeepSeek's compute ownership advantage — teortaxesTex · 2026-09-11