MiniMax has reverted to utilizing the Full Attention architecture. Meanwhile, Kimi has made its Kimi Linear solution open-source. Both companies are pursuing distinct technical routes in their quest for innovation.