What changed
Microsoft Corp. today announced the release of new generative AI models and updates to Foundry, its unified enterprise platform for AI application development and autonomous agents. The updates include an enterprise knowledge graph, learning loops for autonomous agents, and new capabilities for reinforcement learning with organizational data. Microsoft also released Decision-1.1, a new decision model, and expanded its MAI family of first-party AI models.
What this means in practice
The updates to Foundry IQ now provide an enterprise knowledge graph out of the box, offering rich context for application development. Foundry uses existing transparency and observability tools to create learning loops, allowing autonomous agents to establish baselines for quality, safety, latency, and cost. Human annotations, now in preview, enable experts to rate and comment on agent traces, feeding into an improvement loop. The new Decision-1.1 model is designed for fast, cost-effective decision-making in software at scale, offering superior capabilities compared to large language models. It is built on Google’s Gemma 4 26B A4 and includes multimodal support for image analysis. The MAI family now includes MAI-Code-2 and MAI-Code-2-Flash for coding tasks, MAI-Image-2.7 and MAI-Image-2.7-Flash for image generation, and MAI-Cyber-2-Flash for cybersecurity. These models are available on GitHub and in Foundry, with some in preview.
Limitations and availability
The enterprise knowledge graph and learning loops are now available in Foundry. Decision-1.1 is generally available, while MAI-Code-2 and MAI-Image-2.7 are generally available on GitHub, with Code-2-Flash in private preview in Foundry. MAI-Image-2.7-Flash and MAI-Cyber-2-Flash are in preview. Foundry now supports over 11,000 models from various providers, including OpenAI, Anthropic, Meta, Mistral AI, and the open-source community. The continuous improvement workflow for models is integrated with agent improvements.
