AI Agent News/stories/microsoft-adds-decision-1-1-and-new-mai-models-to-foundry-expands-continuous.md

Infrastructure

Microsoft adds Decision-1.1 and new MAI models to Foundry, expands continuous improvement tools

Microsoft has launched new first-party models and updates to Foundry, its platform for building and deploying AI applications and agents. The changes include Decision-1.1, now generally available, plus new MAI coding, image and cybersecurity models.

Source: siliconangle.com/2026/11/17/microsoft-rolls-out-new-first-party-models-and-updates-to-agentic-ai-continuous-improvement

Microsoft says Decision-1.1 is generally available. It is designed to make fast, low-cost decisions in software, such as routing, classification and prioritization, instead of having a large language model reason through them. The model builds on Microsoft-Decision-1, released in October, and is built on Google's Gemma 4 26B A4 base with added multimodal support for images in instructions.

Microsoft also added MAI models to Foundry. MAI-Code-2 is generally available on GitHub for demanding agentic coding work, and MAI-Code-2-Flash targets high-volume everyday coding, with Code-2-Flash in private preview in Foundry. MAI-Image-2.7 and MAI-Image-2.7-Flash are in preview, and MAI-Cyber-2-Flash is in preview within Codename MDASH, Microsoft's agentic code security system.

On the platform side, Foundry IQ now provides an enterprise knowledge graph out of the box. Human annotations, now in preview, let experts rate and comment on agent traces, and Foundry supports reinforcement learning on supported models using organizational data. Foundry's model catalog now features the top 50 models and more than 11,000 models from providers including OpenAI, Anthropic, Meta and Mistral.

Why it matters if you build agents

Teams building agents can use Decision-1.1 for routing and classification steps that currently go through a general LLM, and can act on its calibrated confidence scores to defer when uncertainty is high.

Developers using Foundry can add expert annotations to agent traces and compare evaluation results against real-world outcomes as part of a repeatable improvement loop.

Preview labels matter here: the Flash coding model, image models and cyber model are not yet generally available, so check access before planning production use.

Get stories like this every weekday

One short email with the day's agent tooling news.

Subscribe