According to CoinMeta, Cohere has released a new generation of embedding model, Embed 5, which is divided into two tiers: Pro and Fast. This model is responsible for converting text and images into vectors for use in search, RAG, and agent for information retrieval. Both models support text, image, and mixed inputs, covering over 100 languages with a maximum context length of 128k token. Enterprises can first use the higher-quality Pro to index documents and then use the faster Fast to process queries without the need to regenerate the entire set of vectors. Tests on 40 datasets have shown that the combined performance of Pro for indexing and Fast for querying is equivalent to 98.4% of the full Pro solution, with a difference of about 1.6%. The pricing is $0.08 per million words for token and $0.12 for Pro.