Kimi K2.8 Preview fully launched: performance approaching K3, supporting 1M context

Beating AI News Flash: Moonshot AI has begun rolling out Kimi K2.8 Preview to all Kimi Code CLI users. Users do not need to switch Model IDs — the original kimi-for-coding will point directly to the new model.

Kimi claims that K2.8 Preview's overall performance is already close to K3, but with higher thinking efficiency. It also supports the same thinking intensity adjustment as K3 and expands the context to 1M tokens.

The previous generation K2.7 Code had only 256K context and could only enable thinking. K2.8 has essentially brought two core capabilities from K3 down to the default code model.