Networks of evolutionary processors with an underlying octahedron graph consist of 7 language processors which are linked to the vertices of the octahedron graph. Notice that they are located in the 6 facets and the core of a cube graph. Also note that the nodes are only able to perform a type of mutation based on the words found in that node. Each node is associated with an input filter and an output filter, defined by some regular language. Rules are applied to all the words existing in every node. The words, able to pass the output filter of the respective node, are sent out and they navigate through the graph. Such words will enter those nodes provided their input filters are satisfied. The computational power of the network is comparable to Turing machines when the filters are regular languages. We introduce several variants of octahedron networks, depending on rule types and the way of computation plus their computational power. Some known problems are addressed at the end.