Vibeleaderboard
← All Intel
Intel / article

Decision-1.1 now available on AI Gateway

Source
vercel.com
Author
Kevin Dawkins
Date

Context

Decision-1.1 is an update to Microsoft's decision model, which answers questions about a piece of text with structured results an application can act on instead of free-form prose. A request can ask for the probability that a condition is true, a chosen category with probabilities for every option, or a score on a scale the developer defines, and several questions can share the same input. Vercel's changelog gives uses such as routing a support request to the right team, choosing the next step in a workflow and grading AI responses, the same jobs described when Decision-1 arrived on AI Gateway.

Vercel's changelog, in its updated form, says Decision-1.1 is more accurate, better and faster than Decision-1, and gives no numbers for any of the three. Calibration matters for this kind of model. A calibrated model's probabilities match how often it is right, so a developer can set a threshold, act automatically above it and send cases below it for human review. That only works if 80% means roughly 80%, which is why better calibration would make the probabilities more useful, though how much better is not shown. Switching is a one-line change: Decision-1.1 accepts the same request format, under the model ID microsoft/decision-1.1.

Key takeaways · AI-distilled
  • Microsoft AI's Decision-1.1, now on Vercel AI Gateway, returns structured answers rather than free text: a probability that a condition is true, a chosen category with per-option probabilities, or a score on a scale you define.
  • Several questions can share one input in a single request, and Decision-1.1 takes the same request format as Decision-1, so switching is a model ID change to microsoft/decision-1.1.
  • Vercel's examples need ai 7.0.128 or later in TypeScript or OpenAI Python SDK 3.26.0 or later, and decision requests show in request logs, count toward budgets and support decision fallbacks.
Terms in this piece · Glossary
  • calibration — How well a model's confidence matches reality — a calibrated model saying "90% sure" is right about 90% of the time.
More from Kevin Dawkins
Recommended reads
Comments

Checking sign-in…

Loading comments…