Skip to main content

Microsoft's New AI Model Nails 36 Tests, Challenges GPT-6

Microsoft's New AI Model Nails 36 Tests, Challenges GPT-6

Microsoft has quietly rolled out a new AI model designed for structured decision-making tasks—and it's already turning heads. Called Microsoft-Decision-1, the model has completed field tests across multiple internal scenarios, and its speed is raising eyebrows.

A Decision-Making Powerhouse

In 36 authoritative benchmark tests covering roughly 150,000 questions, Microsoft-Decision-1 consistently ranked first in accuracy. That's not just a win—it's a clean sweep. The model's core strength lies in picking the best option from a predefined set and assigning probability scores to each alternative. Think of it as a super-fast, super-accurate decision engine for business tasks like routing distribution and content classification. It's built to plug into existing apps, AI agents, and workflow systems, making complex processes run smoother.

Under the Hood

The model is built on Qwen3.5-9B and has undergone specialized post-training. Microsoft hints that it plans to migrate the model to other base models down the road. Stability is a strong suit: even after various perturbations, the decision reversal rate stays very low, and it passes multiple strict security tests with flying colors.

The Price Tag That Turns Heads

Here's where it gets interesting. Microsoft-Decision-1 costs just $0.042 per million input tokens, and output is completely free. In real-world tests, it matches the quality of some large models on similar decision-making tasks—but it's faster and drastically cheaper. That's a combination that could make it a go-to for developers watching their budgets.

Where to Get It

Developers can access the model right now through Microsoft's AI development platform. And there's more: Microsoft plans to integrate it into OpenRouter in the future, giving developers even easier access.

Key Points

  • Microsoft-Decision-1 tops 36 benchmark tests covering ~150,000 questions, ranking first in accuracy.
  • It's built on Qwen3.5-9B, with plans to migrate to other base models later.
  • Pricing is aggressive: $0.042 per million input tokens, output free.
  • Available now on Microsoft's AI platform; coming soon to OpenRouter.
  • Ideal for routing, classification, and other structured decision tasks in business workflows.