The paper proposes the Ternary Sentiment Embedding Model, a new model for creating sentiment embeddings based on the Hybrid Ranking Model of Tang et al. (2016), but trained on ternary-labeled data instead of binary-labeled, utilizing sentiment embeddings from datasets made with different distant supervision methods. The model is used as part of a complete Twitter Sentiment Analysis system and empirically compared to existing systems, showing that it outperforms Hybrid Ranking and that the quality of the distant-supervised dataset has a great impact on the quality of the produced sentiment embeddings.