Microsoft Decision-1 joins Vercel AI Gateway for classification and routing

What happened

Microsoft's Decision-1 model is now available through Vercel's AI Gateway, according to an announcement on the Vercel blog. Developers can call it using the model name microsoft/microsoft-decision-1, and access requires nothing more than a standard AI Gateway API key — no separate Azure subscription and no deployment to provision.

The addition is notable because Decision-1 is not a general-purpose text generator. It is built for a narrower job: producing structured decisions about a piece of text rather than open-ended prose.

A model built for structured decisions

As Vercel describes it, Decision-1 responds to typed questions about input text in three forms. Boolean questions return a yes-or-no answer. Choice questions ask the model to pick from a named set of categories. Score questions evaluate an input against an ordered set of criteria.

Each format maps onto a common task in application development. Vercel points to classifying support requests as an obvious case, along with selecting the next step in a workflow and grading another model's response against a rubric.

Two design details stand out. First, multiple questions can be bundled against the same input in a single request, so an application that needs several judgments about one message or document does not have to issue a separate call for each. Second, choice and score answers come back with probabilities attached to every option. That gives the calling application a confidence signal it can act on: high-probability results can be handled automatically, while uncertain ones can be routed to a human for review.

Three ways to call it

Vercel offers three integration paths. The AI SDK decision API is the native option for developers already building on the AI SDK. There is also an OpenAI-compatible Decisions API for code that speaks that protocol, and a TypeSafe-compatible API. All three route through AI Gateway with the same key.

One bill, one set of logs

The gateway angle matters as much as the model itself. Vercel positions AI Gateway as a unified layer for reaching many models while tracking usage and cost in one place. Decision-1 calls behave like any other request in that respect: they appear in gateway logs and count toward configured budgets, so teams adopting the model for classification or routing do not create a second, unmonitored line item in their infrastructure.

Vercel also directs developers to its decision documentation covering the Choice, Score and Boolean question types, alongside a catalog of other decision models available on the gateway.

Why it matters

A large share of real-world LLM plumbing is not about generating text — it is about making small, reliable decisions. Should this ticket go to billing or to technical support? Is this answer good enough to show the user? Which branch of the workflow runs next? Many teams currently handle this by prompting a chat model and parsing whatever comes back, an approach that tends to be fragile, hard to audit and offers no principled measure of confidence.

A model that natively returns typed answers with probabilities changes that calculus. It makes classification and routing feel less like prompt engineering and more like calling a well-specified API, and the built-in confidence values make the escalate-to-a-human pattern straightforward to implement. Bundling it into AI Gateway removes the Azure setup barrier entirely, which should lower the cost of experimentation for teams already in that ecosystem. For anyone building agentic workflows or automated quality checks, this is the kind of quiet infrastructure improvement worth knowing about.