WizardLM-2 8x22B

WizardLM-2 8x22B is a mixture-of-experts instruction model from Microsoft Research, released in April 2024 as a fine-tune of the Mixtral 8x22B base using Microsoft's Evol-Instruct training pipeline. With 141B total parameters and roughly 39B active per forward pass, it briefly led open-weights conversational benchmarks at release and still performs competitively on multi-turn reasoning tasks. The 64K context window is workable for most production retrieval-augmented generation pipelines. Its WizardLM 2 Community License is permissive in practice but carries non-standard attribution clauses, so read it before deploying commercially. By mid-2026, newer MoE fine-tunes have largely passed it on standard benchmarks, but teams already integrated on Mixtral-class infrastructure will find it a low-friction swap at similar cost per active parameter.