Three large open-weights models from 2024, each positioned for a different enterprise buyer. Command R+ from Cohere targets enterprise teams building RAG and tool-use pipelines. At 104B parameters and 131K context, it performs reliably on function-calling evaluations and was designed from the ground up around grounding and retrieval workflows. The CC-BY-NC license means commercial workloads run through Cohere's managed API — self-hosting is not a production option without a separate agreement. If your budget is Cohere's API and you want enterprise support contracts, this is a coherent choice. DBRX Instruct is Databricks' March 2024 MoE, 132B total parameters with 36B active per token and a 32K context window. It outperformed Mixtral 8x7B at launch on MMLU and HumanEval, but the context ceiling and the narrower provider ecosystem have limited its appeal outside the Databricks platform. The Databricks Open Model License covers most commercial use. Since late 2024, models like DeepSeek V3 and Mixtral 8x22B have overtaken it on cost-adjusted quality, making Databricks ecosystem integration the primary remaining justification. Llama 3.1 405B Instruct is Meta's July 2024 maximum-scale open release — 405 billion dense parameters, 131K context, Llama 3 community license. It represents the quality ceiling for openly licensed models from Meta and is the right choice when raw capability at maximum parameter scale is the requirement and you have the infrastructure to run it. Lambda Labs and a handful of GPU-heavy hosts carry it. Per-token costs are significantly higher than 70B alternatives. Pick Command R+ for managed enterprise RAG with Cohere tooling. Pick DBRX when you are already in the Databricks ecosystem. Pick Llama 3.1 405B when you need the highest parameter count with a near-Apache license and have the infrastructure to justify it.