Microsoft Foundry adds GPT-6 and Claude 5.5 for enterprise AI agents

The inclusion of GPT-6 and Claude 5.5 signals a shift where enterprise value is derived from the orchestration layer rather than a single model provider. By providing a unified framework for voice, long-running tasks, and tool discovery, Microsoft is positioning Azure as the essential middleware for complex streaming and media workflows that require high-reliability AI. This move pressures competitors to offer similar cross-provider flexibility to prevent developer lock-in. Watch for the October launch of the AI Gateway tier to see how centralized governance impacts the speed of agent deployment in highly regulated media environments.

Microsoft has been building Foundry Agent Service toward production readiness throughout 2026. At Build 2026 in May, Microsoft announced that hosted agents in Foundry Agent Service would reach general availability by early July 2026, with sandboxed sessions, durable filesystem access, and framework flexibility that lets agents built with Microsoft Agent Framework, GitHub Copilot SDK, LangGraph, or other SDKs deploy without rewrites. The runtime supports both the Responses API for OpenAI-compatible stateful interactions and the Invocations protocol for schema-free pass-through scenarios, giving developers two distinct integration paths depending on their orchestration needs.

The model-agnostic positioning extends beyond the September GPT-6 and Claude Opus 5.5 additions. Microsoft's March 2026 Foundry update made the next-generation Foundry Agent Service generally available, built on the OpenAI Responses API and wire-compatible with models from DeepSeek, xAI, Meta, LangChain, and LangGraph. The same update introduced Voice Live as a fully managed real-time speech-to-speech runtime that collapses the traditional STT, LLM, and TTS pipeline into a single managed API, incorporating semantic voice activity detection, end-of-turn detection, server-side noise suppression, echo cancellation, and barge-in support.

For teams evaluating agent optimization workflows, Microsoft shipped Agent Optimizer alongside the September announcement. The tool evaluates agents across multiple model deployments in the same run, scoring each option against custom evaluators, allowing developers to determine whether a smaller model handles their workload as well as a larger one or whether stepping up to a higher-tier model produces meaningful quality improvements on the dimensions that matter for their specific agent. This continuous optimization loop, combined with the platform's tracing and evaluation capabilities, positions Foundry as a closed-loop system that converts production failures into ranked, reviewable agent improvements.