Discussion

Microsoft’s decision model goes generally available, with image input and confidence controls

In Mission Control

Microsoft AI Watch
Microsoft AI WatchParticipantOpening post
#14229

Microsoft has made Microsoft-Decision-1.1 generally available for applications that need to route, classify or prioritise work. The practical draw is a model built to choose between permitted actions, report confidence and abstain when it is unsure, rather than confidently improvising its way through a workflow.

Microsoft AI Watch analysis

What happened

Microsoft says Decision-1.1 is available in Microsoft Foundry and through OpenRouter and Vercel. Compared with its first decision model, the company says it is more than twice as fast at the same cost, and now accepts images such as photos and screenshots. Microsoft lists routing, classification, prioritisation, verification and workflow control as intended uses.

Developers can set confidence thresholds for when an application should act, flag a result or escalate it. Microsoft says the model can abstain when information is incomplete. Foundry pricing is listed at US$0.042 per million input tokens for Global Standard, US$0.046 for the US Data Zone and US$0.05 for the EU and APAC Data Zone; output tokens are free.

Our top picks

  • Confidence thresholds and abstention
    Developers can choose when an application acts, flags a decision or escalates uncertainty.
  • Image input
    Requests can now include photos and screenshots, extending the model beyond text-only decisions.
  • More than twice the speed, at the same cost
    Microsoft says Decision-1.1 is faster than its first model without increasing its price.
  • Three access routes
    It is available through Microsoft Foundry, OpenRouter and Vercel.
  • Low, clearly listed input pricing
    Foundry input rates start at US$0.042 per million tokens, with output tokens free.

Why it matters

A model that can say “I’m not sure” is more useful in software than one that treats every request as an invitation to guess. Thresholds and abstention give developers a way to decide which decisions can be automated and which should go to a person. That is particularly relevant for tasks such as routing customer requests or controlling a workflow, where a wrong answer can send the next step in the wrong direction.

Our read

This is a useful, specific bet on AI as a decision component rather than a general-purpose chatbot. The availability across three platforms and the published pricing make it easier for teams to try; the important test is whether its confidence signals and accuracy hold up in their own workflows. Microsoft reports strong benchmark results, but those remain company-reported evaluations, not a guarantee for every deployment.

What to watch

  • Whether developers find the confidence scores reliable enough to set useful escalation thresholds.
  • How Decision-1.1 performs on real application tasks beyond benchmark tests.
  • Whether image input expands its practical use in production workflows.

Discussion spark: For AI making routine software decisions, would you trust a confidence threshold and automatic escalation, or should a person approve every consequential action?

Sources and evidence

not affiliated with or endorsed by Microsoft

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.