Llama 3 GA in Azure Databricks Foundation Model APIs
meta databricks azure llama model-release enterprise-ai
trending_up Trend: meta

Llama 3 GA in Azure Databricks Foundation Model APIs

calendar_month July 3, 2026

Summary

Microsoft and Databricks have announced the general availability (GA) of Meta Llama 3 models (8B and 70B) within the Azure Databricks Foundation Model APIs. This native integration enables developers to query and deploy Llama 3 models directly within their Databricks workspace without managing dedicated GPU infrastructure. Both pay-per-token and provisioned throughput billing options are available. The feature ensures enterprise-grade security and compliance by keeping data entirely within the Databricks environment under Unity Catalog governance.

What happened?

As part of their ongoing AI platform expansion, Microsoft and Databricks integrated Meta’s Llama 3 models natively into the Azure Databricks Foundation Model APIs. The models, available in 8-billion (8B) and 70-billion (70B) parameter sizes, are now ready for production workloads. Developers can access them using standardized, OpenAI-compatible REST APIs.

Why it matters

  1. Zero Infrastructure Overhead: Developers no longer need to provision, configure, or scale GPU-based virtual machines or Kubernetes clusters to host Llama 3. Databricks manages the backend hosting and scaling automatically.
  2. Data Security and Compliance: Enterprise customers are often hesitant to send sensitive business data to external APIs. With this integration, all data remains entirely inside the secure Databricks environment under the governance of the Azure Unity Catalog.
  3. Flexible Cost Structure: The pay-per-token tier allows companies to pay only for the exact volume of tokens processed. For high-volume production applications requiring guaranteed latency, the “Provisioned Throughput” option provides dedicated capacity.

Evidence

The official Azure Updates portal (Update ID 567194) confirmed the general availability of Llama 3 models in the Azure Databricks Foundation Model APIs. Additionally, Databricks and Microsoft technical documentation details the availability of these models under Databricks Model Serving.

Analysis

This release highlights a clear trend in enterprise AI: moving away from self-hosting open-source models toward fully managed, serverless endpoints. Hosting a large model like Llama 3 70B on private clusters requires significant maintenance and optimization. Cloud providers are standardizing this workflow. Databricks is positioning itself as a unified data platform that brings LLMs directly to where the data lives, reducing latency and mitigating security risks.

Practical Takeaways

  • For Developers: Llama 3 models can be queried immediately using Databricks SDKs or standard REST APIs with minimal setup.
  • Cost Optimization: Use the pay-per-token model for prototyping, testing, or low-frequency batch jobs. Transition to provisioned throughput for high-scale, latency-sensitive production workloads.
  • Data Governance: Since Unity Catalog governs these Foundation Model APIs, organizations can apply access control policies and audit trails directly to LLM interactions.

Open Questions

  • How will future model iterations (like Llama 3.1 or newer releases) affect the compatibility and pricing of current Llama 3 endpoints?
  • What are the precise latency differences between native Databricks serving and alternative Azure hosting methods, such as Azure AI Studio?

Sources

  1. Azure Updates: Llama 3 in Azure Databricks Foundation Model APIs
  2. Databricks Documentation: Foundation Model APIs