This work presents a new class of reaction automata, called Chemical Pure Reaction Automata (CPRA). CPRA combines characteristics of chemical reaction automata, as introduced by Okubo et al. in 2016, with those of the more recently defined pure reaction automata. Unlike standard chemical reaction automata, CPRA lack permanence, meaning their result states consist solely of the reaction products, with unconsumed reactants being discarded. We investigate the computational power of two CPRA variants, both working in the maximally parallel manner. We first prove that deterministic CPRA (DCPRA) -- in which at every state, for each input symbol, the result state is the same for all multisets of enabled reactions -- are not Turing complete. We then show that non-deterministic CPRA are Turing complete and thus strictly more powerful than DCPRA: namely, the set of languages accepted by CPRA in the maximally parallel manner contains the set of languages accepted by standard chemical reaction automata in the same manner.
