Microsoft Releases Microsoft-Decision-1, a Qwen3.5-9B Scoring Model

Microsoft has released Microsoft-Decision-1, a specialized decision-scoring model designed for routing, classification, verification, and agent control tasks. According to a report by MarkTechPost, the model is post-trained from Alibaba’s Qwen3.5-9B and is available immediately on Microsoft Foundry and OpenRouter.

Unlike standard large language models that generate free-form text, Microsoft-Decision-1 returns a calibrated probability for each fixed answer option. This design allows developers to integrate precise decision-making capabilities into applications without parsing unstructured text outputs. The model focuses on scoring specific choices rather than creating narrative content.

View the original post on X

Model Architecture and Functionality

Microsoft-Decision-1 is built upon the foundation of Alibaba’s Qwen3.5-9B model. The company post-trained this base model to specialize in decision-scoring tasks. According to MarkTechPost, the resulting system provides calibrated probabilities for predefined answer sets. This approach differs from generative AI models that produce variable text responses.

The model is intended for use cases requiring clear binary or multi-class decisions. These include routing queries, classifying data inputs, verifying information, and controlling autonomous agents. By outputting probabilities for fixed options, the model reduces ambiguity in automated decision pipelines.

Availability and Access

Developers can access Microsoft-Decision-1 through two primary channels: Microsoft Foundry and OpenRouter. The release makes the model available for immediate integration into existing workflows. MarkTechPost notes that the model is live on both platforms.

Microsoft Foundry serves as the company’s platform for building and deploying AI applications. OpenRouter provides a unified API for accessing various AI models. This dual availability allows developers to choose their preferred infrastructure for integrating the decision-scoring capabilities.

What should you do?

Security and technology professionals should evaluate whether their applications require deterministic decision outputs rather than generative text. Teams building agent-based systems should test Microsoft-Decision-1 for routing and verification tasks. Developers should review the calibration accuracy of probability scores in their specific use cases.

Background: Large language models

Large language models are a specific type of neural network trained on text. Generative AI is the broader umbrella term for any system that creates new content, including images and code. Not all generative AI uses large language models, but most modern text generators do.

Read the full guide: Large Language Models vs Generative AI: The Distinction and Security Implications

What to do and how to stay safe: Microsoft

• Audit current applications to identify tasks that require fixed-choice outputs instead of generative text responses.
• Review the documentation for Microsoft-Decision-1 on Microsoft Foundry and OpenRouter to understand input format requirements.
• Test the model’s probability calibration against your specific classification or routing datasets to ensure accuracy.
• Monitor for any updates to the model’s training data or performance benchmarks from Microsoft or Alibaba.

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is Microsoft-Decision-1?

Microsoft-Decision-1 is a decision-scoring model released by Microsoft that returns calibrated probabilities for fixed answer options instead of generating text.

What is the base model for Microsoft-Decision-1?

The model is post-trained from Alibaba’s Qwen3.5-9B, according to a report by MarkTechPost.

Where is Microsoft-Decision-1 available?

The model is available now on Microsoft Foundry and OpenRouter.