In recent years the use of social media platforms in Indonesia has continued to increase. The increasing use of social media has several advantages and disadvantages. The advantage is that the news is easily accessible by anyone, while the disadvantage is that much information that is spread is hoax news. Hoax news must be detected because hoax news spreads false and misleading information. This undermines the integrity of the information and needs to be clarified for the public. By detecting hoax news, we can ensure the information being disseminated is accurate and trustworthy. In this study, the author will detect hoax news on Indonesian news media on Twitter using LSTM with word embedding GloVe and Word2Vec and compare the two-word embeddings to find the best performance in the LSTM model. The reason for choosing the GloVe and Word2Vec extraction features to be compared is that both are useful for representing vectors of words. Their performance may vary. Word2Vec might better capture semantic relationships between words, whereas GloVe might better capture distributional relationships and word co-occurrence. This study shows that LSTM with Word2Vec performs better than LSTM and GloVe in detecting Indonesian language news. LSTM and Word2Vec produced an average accuracy value of 95%, while LSTM with GloVe produced an average accuracy value of 90%.