Blended Skill Talk is a dataset about conversations. The dataset analyzes how features blend in natural conversations and compares the performance of different architectures and training schemes, which are important capabilities for conversational agents. A good open-domain conversational agent is able to blend qualities such as attractiveness, knowledgeableness, and empathy in conversations. The dataset compares the performance of different architectures and schemes, from simple model aggregation schemes, which require minimal additional training, to various forms of multi-task training. Experiments show that multi-task processing schemes can produce better conversational performance than models trained on a single task.