Jet reconstruction, identification and classification is of prime interest in an hadronic environment such as the LHC. Algorithms were developed in order to separate jets emerging from the decay of charm and bottom quarks, and to identify large jets produced from the decays of heavy resonances. In Run 2, such algorithms have benefited from the use of increasingly complex deep neural network architectures. Improved performance are now allowing to probe physics processes thought to be unreachable at hadron colliders. In parallel, new methods have been investigated to derive corrections to apply to the simulation to match the data taken at the LHC.