Mixtral 8x7B: What you need to know about Mistral AI’s latest model

Mistral AI released Mixtral 8x7B, a Mixture-of-Experts LLM that matches or outperforms GPT-3.5 on most benchmarks while supporting 32k tokens and improved code generation. The post covers benchmark comparisons against LLaMA 2 70B and GPT-3.5, hardware requirements for self-hosting (64GB RAM, 2 GPUs), the Mistral AI API with Python and JavaScript SDKs, the new mistral-embed 1024-dimension embedding model, and a pricing comparison showing mistral-tiny and mistral-small are cheaper than gpt-3.5-turbo. Developers using pgvector with ada v2 embeddings will need to re-create embeddings to switch to mistral-embed.