Microsoft Foundry Agent Service now lets enterprises choose more AI models

Image via Depositphotos.com

Microsoft Foundry Agent Service, a platform designed to help enterprises build AI agents, has been updated to support more models, including those on the frontier, or ones that are more specialist. The new models come from several partners including Anthropic, DeepSeek AI, Meta, Microsoft, and xAI. With this update, enterprises can build secure AI agents without being locked into a single provider.

The expanded catalogue of models allows developers to pick the right model for their needs; some excel at reasoning depth, while others are designed for context length, multimodality, and cost. Newly supported models that excel in frontier reasoning including DeepSeek-R1-0528 (advanced long-form and multi-step reasoning), DeepSeek-V3-0324 (multimodal understanding across text and images), grok-4 (frontier-scale reasoning for complex problem solving), grok-4-fast-reasoning (accelerated agentic reasoning optimized for workflow automation).

There are also new Anthropic models available:

  • Claude-Opus-4-1: Frontier reasoning for the most complex problem-solving.
  • Claude-Sonnet-4-5: Balanced performance for multimodal and agentic workflows.
  • Claude-Haiku-4-5: Lightweight, high-speed generation for interactive scenarios.

On the general purpose and enterprise orchestration side, you now have access to Llama-4-Maverick-17B-128E-Instruct-FP8 (optimized for fast, cost-effective inference) and gpt-oss-120b (open ecosystem model supporting transparency and reproducibility). Finally, there are some other xAI models including grok-4-fast-non-reasoning (high throughput, low-latency generation and system routing) and grok-3 (strong reasoning for complex, system-level workflows).

Microsoft"s Foundry Agent Service ensures that these models are production ready by providing streaming responses for real-time engagement, flexible tool calling, multimodal capabilities, grounded retrieval, and model routing that auto selects the best model for the task based on performance, accuracy, and cost.

The new models are available via the SDK, API, and the Foundry Agent Playground. Some models will require registration depending on your region or compliance requirements.

Report a problem with article
Next Article

Germany gets closer to mandating age assurance at OS level

Previous Article

Australia adds Twitch to its teen social media ban