Llama 4 Maverick 17B 128E Instruct FP8

Llama 4 Maverick is Meta's multimodal entry in the Llama 4 family, delivered here in an FP8-quantized instruction-tuned variant that retains the visual reasoning capabilities introduced with the broader collection. Microsoft's catalog characterization frames the model as strong at precise image understanding and creative writing, positioning it as a higher-quality, lower-priced alternative to the earlier Llama 3.3 70B. The Hugging Face repository confirms the weights are distributed under the Llama 4 Community License Agreement (Effective Date April 5, 2025), with Meta's official documentation hosted at llama.com/docs/overview, giving teams a clearly licensed, openly shared base to build on rather than a closed proprietary endpoint.

Beyond its text-and-image reasoning profile, the model is designed to fit production multimodal workflows where cost and latency matter more than flagship-scale reasoning. The Foundry listing treats it as a "Direct from Azure" enterprise offering with unified billing and governance, while the open-weights release under Meta's community license lets the same underlying model travel across clouds, including through hosted catalogs that surface attachments, tool calling, and temperature control. In practical terms, Maverick is well suited to vision-grounded assistants, document and chart Q&A, image captioning pipelines, and creative content tasks where pairing tight per-token economics with reliable image comprehension is the deciding factor.