Llama 4 Maverick 17B 128E Instruct FP8

Llama 4 Maverick is part of the Llama 4 family released by Meta in early April 2025, with the FP8 Instruct variant published on Hugging Face under the Llama 4 Community License Agreement. On Azure AI Foundry, it is offered as a Direct from Azure first-party model, deployable as a serverless endpoint and provisionable through ARM templates. This combination of Meta-published weights and Azure-managed hosting gives teams a path to use a frontier-tier open-style Llama model without standing up their own inference infrastructure, while still having the option to run the open weights outside Azure when desired.

The model fits well for teams that want Llama 4 Maverick's general-purpose chat and reasoning behavior inside Azure's managed environment, including custom endpoint deployments for production applications. Its availability as a Direct from Azure SKU means it can be integrated alongside other Azure AI services, with the underlying open weights usable both on the managed endpoint and in self-hosted setups where licensing terms allow. Organizations already standardized on Azure can use it as a drop-in option for building assistants, multi-step workflows, and other text-generation tasks, while keeping the freedom to switch to self-hosted deployments if cost, data-residency, or fine-tuning needs change.