Mistral AI released Mixtral 8x7B, a sparse mixture-of-experts model with eight experts: roughly 46.7 billion parameters in total but only about 12.9 billion used per token. It beat Llama 2 70B on most benchmarks with six times faster inference, and shipped under the same Apache 2.0 licence as Mistral 7B. The significance was that a technique the large labs had been using without describing was now available as weights anyone could download; Google would not confirm MoE in a flagship model until Gemini 1.5 Pro in February 2024. The same day, Mistral disclosed a $415 million (€385 million) Series A led by Andreessen Horowitz at a valuation of about $2 billion, and opened a commercial platform serving both Mistral 7B and Mixtral through an API.