Mixtral 8x22B Instruct
Mistral AI · 2024-04-17 · 140.6B parameters
Mixtral 8x22B Instruct v0.1 is Mistral AI's instruction-fine-tuned checkpoint of the sparse mixture-of-experts Mixtral 8x22B, released on April 17, 2024. Its 56-layer configuration routes each token through two of eight local experts, supports 65,536-token sequences and native function calling, and Mistral reports 141B total parameters with 39B active alongside fluency in English, French, Italian, German, and Spanish and strength in mathematics and coding. The Apache-2.0 checkpoint remains downloadable, although Mistral retired its `open-mixtral-8x22b` API model on March 30, 2025.