Artificial intelligence is becoming an increasingly important part of the applications and services we use every day. Whether it is Microsoft Copilot helping us work with information, AI agents automating business processes, or applications analyzing large amounts of data, we are seeing more organizations integrate AI into their everyday operations. However, not every application needs an AI model that can write detailed responses, generate documents, or hold conversations. Sometimes, all we really need is a model that can evaluate information, make a decision, and tell an application what should happen next.

On November 17, 2026, Microsoft announced the general availability of Microsoft-Decision-1.1, a new AI model designed to make these types of decisions quickly, consistently, and at a relatively low cost. Rather than focusing on generating content like a traditional large language model (LLM), Microsoft-Decision-1.1 helps applications choose between options, categorize information, prioritize tasks, and determine the next appropriate action. This might sound like a small distinction, but it could make a significant difference in how organizations build and use AI-powered applications.

The model is now available through Microsoft Foundry, OpenRouter, and Vercel, making it accessible to developers who want to introduce intelligent decision-making into their applications without relying on larger, more expensive AI models for every task.

What Exactly Is Microsoft-Decision-1.1?

To understand what Microsoft-Decision-1.1 does, it helps to think about how we typically use AI today. When we ask Microsoft Copilot or ChatGPT a question, the AI processes the request and generates a response. That response might be a few sentences, a detailed explanation, or even a complete document. This is useful when we need the AI to interpret information, explain something, or create new content, but it is not always the most efficient way to make a simple decision within an application.

Consider an IT service desk application that receives hundreds of support requests each day. The application needs to determine whether each request is related to networking, identity, security, or a general technical problem. It might also need to decide how urgent the request is and which support team should receive it. A traditional AI model could analyze each request and generate an explanation, but the application does not necessarily need that explanation. It simply needs to know which category applies, how important the request is, and where to send it.

This is where Microsoft-Decision-1.1 comes in. Instead of producing a lengthy written response, the model evaluates the information provided and selects from a set of predefined options. It also provides confidence information that helps the application determine how certain the model is about its choice. Developers can use that information to decide whether the application should proceed automatically or request additional review.

The important difference is that Microsoft-Decision-1.1 is designed to help applications make structured decisions rather than generate conversations or content. This makes it particularly useful for repetitive activities where applications need to evaluate information quickly and consistently, sometimes thousands or millions of times.

How Does It Work?

The basic idea behind Microsoft-Decision-1.1 is relatively straightforward. An application provides information about a particular situation, along with the possible choices that the model is allowed to make. The model evaluates that information and returns a structured result that the application can use to determine what happens next.

For example, imagine an organization has built an application to manage incoming security alerts. When an alert arrives, the application could provide the model with information such as the alert type, severity, affected system, and whether similar alerts have occurred previously. The model could then evaluate that information and recommend whether to classify the alert as low, medium, or high priority. Based on the result, the application could assign the alert to a particular team, start a predefined workflow, or flag it for further investigation.

This approach is useful because the model does not need to invent its own course of action. The application defines the available choices, and the model evaluates which option is most appropriate based on the information it receives. Developers can also set confidence thresholds so uncertain decisions are reviewed rather than acted on automatically.

For example, if the model is highly confident that a support request belongs to the networking team, the application might automatically assign the ticket. However, if the model cannot confidently determine the appropriate team, the application could forward the request to a person for review. This helps organizations use AI automation without assuming that every AI-generated decision will always be correct.

Of course, the model’s confidence score is not a guarantee of accuracy, and organizations still need to test how well it performs against their own data and business requirements. However, having this information available gives developers another way to introduce controls into automated workflows.

Where Could Organizations Use It?

One of the most interesting aspects of Microsoft-Decision-1.1 is that it is not limited to a particular industry or application type. Any system that regularly needs to evaluate information and choose between different actions could potentially benefit from a decision model. Microsoft has identified several scenarios where the technology can be useful, including request routing, information classification, prioritization, verification, and automated workflow control.

For organizations already using Microsoft technologies, there are several practical examples worth considering:

ScenarioHow Microsoft-Decision-1.1 could help
IT service managementCategorize support tickets and assign them to appropriate teams
Security operationsHelp prioritize alerts and determine which require further investigation
Business process automationSelect the next step in an approval or processing workflow
Document managementClassify documents based on their content or intended purpose
Customer supportIdentify the type of request and direct it to the correct service
AI agentsDecide which tool, model, or predefined action should be used next
Quality assuranceEvaluate whether an AI response meets defined requirements

These examples demonstrate how a relatively small decision can play an important role within a larger application. An organization might already have an automated workflow that receives requests, processes information, and sends notifications. Introducing a decision model could make that workflow more flexible by allowing it to evaluate different situations instead of relying entirely on traditional rules.

This does not mean decision models should replace every existing rule or automated process. If a decision can be made reliably using a straightforward condition, such as checking whether a value exceeds a predefined limit, traditional application logic may still be the simplest and most appropriate solution. Decision models become more interesting when the information is less predictable, such as interpreting the meaning of a support request or determining which category best describes a document.

Why Is Microsoft Focusing on Decision Models?

As more organizations introduce AI into their applications, the cost and performance of those models become increasingly important. A large language model may be extremely capable, but using one every time an application needs to make a simple decision can introduce unnecessary processing time and expense. This becomes particularly noticeable when applications handle large numbers of requests or when several AI models work together to complete a business process.

Microsoft-Decision-1.1 is designed to address this by focusing specifically on decision-making tasks. According to Microsoft’s announcement, the new version is more than twice as fast as the original Microsoft-Decision-1 model, while maintaining the same pricing. Microsoft has also improved the model’s ability to assess uncertainty, which matters when applications need to decide whether a decision should be trusted or escalated.

Another interesting improvement is the addition of image support. This means the model can evaluate information that includes photographs or screenshots, rather than being limited to text-based inputs. For example, an application might use a screenshot to determine which predefined action to take next, or evaluate an image as part of a classification process. This introduces additional possibilities for applications that interact with visual information, although the model is still intended to produce structured decisions rather than detailed image descriptions.

Microsoft already uses the technology in its own products. For example, Microsoft-Decision-1.1 helps determine whether a user request should be handled by Copilot or Cowork, while other implementations help select the appropriate amount of reasoning for a task. It is also used in Microsoft Foundry Evaluations to assess AI-generated results and in internal support scenarios to identify relevant information during operational incidents.

These examples highlight something important about the future of AI applications. Rather than expecting one large model to handle everything, we are increasingly likely to see applications that combine different models, each designed for a particular purpose. A decision model might determine what needs to happen next, while a large language model handles complex reasoning or generates the final response. This lets developers use more expensive AI capabilities where they add the most value, while simpler decisions are handled more efficiently.

Getting Started with Microsoft-Decision-1.1

For developers interested in exploring Microsoft-Decision-1.1, the model is now generally available through Microsoft Foundry, as well as OpenRouter and Vercel. Microsoft Foundry provides an environment where developers can access AI models, integrate them into applications, and evaluate their performance across different scenarios.

One advantage of Microsoft Foundry is that organizations can experiment with a model before introducing it into their applications. A developer could begin with a simple example, such as categorizing incoming requests, and evaluate how accurately the model selects between different options. Once the results are understood, the same approach could be expanded into a more complete workflow that includes automated actions, confidence checks, and human approval where necessary.

Microsoft has also made the pricing relatively inexpensive for this type of workload. At general availability, the Global Standard price in Microsoft Foundry is $0.042 per million input tokens, with output tokens provided at no additional charge. Pricing varies slightly by selected data zone, so organizations should review the available deployment options and pricing before implementing the model at scale.

For anyone wanting to explore the model, the following resources provide a useful starting point:

Final Thoughts

I think Microsoft-Decision-1.1 is an interesting development because it addresses a part of artificial intelligence that does not always receive as much attention as content generation or conversational AI. Much of the discussion around AI focuses on what models can write, create, or explain, but there is also considerable value in helping applications make better decisions. Whether that involves categorizing a support request, prioritizing a security alert, or determining which action an automated workflow should take next, these are activities that happen constantly within business applications and services.

What I particularly like about this approach is that it encourages developers to think more carefully about where AI is genuinely needed. Not every task requires a large language model, and not every decision needs a detailed explanation. Sometimes, the most useful result is simply an accurate choice between several known options, supported by enough confidence information to determine whether the application should proceed or request further review.

Of course, organizations should still be careful about how much responsibility they give to automated decision-making. A model that works well for routine classification may not be appropriate for making sensitive or business-critical decisions without additional controls. Developers need to understand the consequences of incorrect decisions, test the model against realistic scenarios, and ensure that appropriate safeguards are in place before allowing automated actions.

Ultimately, I see Microsoft-Decision-1.1 as another useful component in the growing collection of AI technologies available to developers. It is not intended to replace traditional AI models or existing application logic, but rather to complement them by handling a specific type of task more efficiently. As organizations continue building AI-powered applications and automated workflows, having a model designed specifically for fast, structured decision-making could help make those solutions more responsive, affordable, and practical to operate.

The most interesting part will be seeing how developers incorporate these capabilities into everyday applications, particularly where intelligent decisions can reduce manual work without adding unnecessary complexity.