Microsoft has expanded Foundry Agent Service with a wide range of AI models. LLMs from Anthropic, DeepSeek AI, Meta, Microsoft, and xAI are now available.
The expanded catalog allows developers to choose the right model for their specific application. The different LLMs excel in a variety of ways, from coding to reasoning based on long inputs. Examples include DeepSeek-R1 (version 0528) and -V3 (0324), as well as variants of Grok-4.
This brings Microsoft’s AI service a step closer to the model selection of Vertex AI Agent Service, Google’s counterpart to Microsoft’s offering. Apart from Google’s own models, which are exclusive to Google Cloud, Google offers mostly the same LLMs. The big absentee is OpenAI, which is only available from Microsoft when it comes to closed-source models. These include the latest variants (GPT-5.1) and legacy options such as o4-mini, o3, GPT-4.1, o1 and (for now) GPT-4o and GPT-4.
Anthropic’s Claude series available
Anthropic’s latest models have also been added. Claude-Opus-4-1 is intended for the most complex issues. Claude-Sonnet-4-5 offers a balance between performance and multimodal workflows. Claude-Haiku-4-5 focuses on speed for interactive scenarios.
General models are available for enterprise applications. Llama-4-Maverick-17B-128E-Instruct-FP8 optimizes for fast, cost-effective inferencing. OpenAI, currently not a big proponent of open-source models, does provide gpt-oss-120b via the Foundry service.
Microsoft is trying to make it as easy as possible to deploy the models. The service offers streaming responses for real-time interactions, flexible tool calling, and multimodal capabilities. Grounded retrieval and automatic model routing select the best model based on performance, accuracy, and cost.
The models are available via the SDK, API, and Foundry Agent Playground.
Read also: Vertex AI expands with Agent Development Kit, Agent Garden, and Agent Engine