Currently, the ChEBI lookup is based on SMILES strings. While we try to canonicalise the SMILES, the canonicalisation still depends on the software used. InChI (or InChI keys) are more commonly used as identifiers and ensure that identical molecules get the same key.