Per-model inference outputs from SLMs and LLMs evaluated on customer service client-agent conversations.