title: "Cheaper, Better, Faster, Stronger | Mistral AI"

Mistral AI has released Mixtral 8x22B, a new open-source model available under the Apache 2.0 license. It is a sparse Mixture-of-Experts (SMoE) model that offers high efficiency by using only 39B active parameters out of a total 141B, delivering a strong performance-to-cost ratio. The model demonstrates superior performance on reasoning, knowledge, math, and coding benchmarks compared to other leading open models. It is also natively multilingual, supports function calling, and has a 64K token context window for recalling information from large documents.