Evaluating reasoning large language models with human-like thinking in ophthalmic question answering | Synapse