Meta’s Llama 4 Scout and Maverick Debut on Azure AI Foundry and Databricks

Details

• Meta's Llama 4 Scout and Maverick models are now available on Microsoft’s Azure AI Foundry and Azure Databricks, enhancing Azure's AI model roster.
• Scout models come with 17 billion parameters and offer a 10 million-token context window for large-scale enterprise data analysis, while Maverick leverages a 128-expert MoE architecture for advanced multilingual chat and FP8 precision tasks.
• The models introduce early-fusion multimodal capabilities and a mixture-of-experts architecture, enabling unified processing of text, images, and video, while delivering 400 billion total parameters with optimized resource consumption.
• This rollout builds on Microsoft’s October 2024 AI Foundry launch, extending capabilities that surpass Llama 3’s 128K-token context limit and achieving stronger performance in reasoning and coding benchmarks.
• Llama 4 models are paired with Azure's integrated safety systems and Meta's adversarial mitigation tools, supporting secure use cases such as automated document analysis and multilingual customer service in enterprise settings.

Impact

The addition of Llama 4 models strengthens Azure’s position in enterprise multimodal AI, directly challenging offerings like OpenAI's GPT-4 and Anthropic's Claude 3. This partnership offers enterprises efficient, customizable solutions, especially for regulated sectors. The shift to scalable, MoE-based architectures is likely to accelerate AI agent adoption across a range of advanced business applications.