WizardLM-2 8x22B

WizardLM-2 8x22B is an open-weight instruction-tuned large language model released by Microsoft AI, built as a fine-tune of Mixtral 8x22B with a Mixture-of-Experts architecture totaling ~141B parameters. It targets general-purpose chat, reasoning, multilingual tasks, and agentic use cases with a context window of approximately 65K tokens. Source: supabase.