Observed facts: Vercel’s changelog says Decision-1 answers yes/no questions, chooses among named categories, or scores inputs against ordered criteria. It also says choice and score answers include probabilities for each option, and multiple questions can share one input. Mechanism: Rather than asking a model for an unconstrained passage and parsing it afterward, an application specifies a question type and criteria, then receives structured answers with probabilities. The application can use those probabilities to act or send uncertain outcomes for review. The changelog lists AI SDK, OpenAI-compatible Decisions, and TypeSafe-compatible APIs, and says no separate Azure account or deployment is required.
Why now: The availability change puts a decision-oriented interface alongside other model requests in AI Gateway, where calls also appear in logs and count toward budgets. First implication: Teams can test bounded classification and rubric-scoring steps within an existing Gateway integration, while retaining an option to route uncertain cases to human review. Evidence boundary: The supplied source establishes availability, interfaces, output behavior, and Gateway accounting; it does not establish comparative accuracy, latency, cost, or production outcomes.
What to watch: Whether Vercel or Microsoft publishes task-level evaluations and operational results for Decision-1 deployments through the Gateway. New branch reason: No prior New Runtime record was supplied for this specific availability change, so this records the distinct integration event without asserting a relationship to another record.