According to CoinMeta, Kimi has announced the full launch of K2.8 and Preview. Their performance is now close to that of K3 and they support up to 1M contexts. Users do not need to switch between model and id; the existing kimi-for-coding will directly point to the new models. Kimi indicates that the comprehensive performance of K2.8 and Preview is close to that of K3, but they offer higher thinking efficiency. They also support the same level of thinking intensity adjustment as K3 and can expand the context to 1M token. In contrast, the previous generation of K2.7 and code only supported 256K contexts and could only enable the thinking function. This time, K2.8 has basically incorporated two core capabilities of K3 into the default code model.