Three enterprise-branded MoE models from companies whose primary business is not large language models. Snowflake Arctic Instruct is a 480B-total-parameter model with roughly 17B active per token, released April 2024 under Apache 2.0. The hybrid dense-plus-sparse architecture was built specifically for enterprise SQL generation and RAG over structured data, where early benchmarks showed competitive performance relative to similarly priced alternatives. The hard constraint is the 4K context window, which rules out most long-document workflows. Hosted coverage is sparse as of 2026; Arctic is primarily a research reference or fine-tuning base for teams inside the Snowflake ecosystem. DBRX Instruct from Databricks launched one month earlier, March 2024, with 132B total parameters and 36B active per token across 16 fine-grained experts. It outperformed Mixtral 8x7B at the time on MMLU and HumanEval, and the 32K context window is a modest improvement over Arctic. The Databricks Open Model License is permissive but not OSI-approved. Since mid-2024, newer models have largely passed it on cost-adjusted quality; the remaining justification is teams embedded in the Databricks platform. WizardLM-2 8x22B is Microsoft Research's April 2024 Evol-Instruct fine-tune of Mixtral 8x22B, with 141B total parameters, 39B active per pass, and a 64K context window. Its benchmark performance on multi-turn conversational tasks exceeded both competitors at launch and still holds up reasonably well. The WizardLM 2 Community License carries attribution clauses that require review before commercial deployment. Pick Arctic for structured-data SQL generation workloads inside the Snowflake ecosystem with Apache 2.0 licensing. Pick DBRX when Databricks integration is the deciding factor. Pick WizardLM-2 8x22B for multi-turn conversational use cases at MoE cost efficiency where licensing terms can be accommodated.