Microsoft Decision 1.1 is generally available: a fast model for yes/no and pick-one choices
Microsoft's Decision-1.1 model, which scores fixed answer options instead of writing text, is now generally available in Foundry, OpenRouter and Vercel, with image input and twice the speed.
- Author
Umair Mateen- Published
- Reading time
- 3 min
- Category
- AI
Ideas in this post
Tap one for a 5-second answer. Underlined words in the article work the same way.
0 of 3 understood

Microsoft has made Microsoft-Decision-1.1 generally available, announced on 17 November 2026. It is a small AI model built only to choose between fixed answers, and it is now offered in Microsoft Foundry and through OpenRouter and Vercel. means it is supported for production use, not just testing.
What it actually does
Most AI models write their answer word by word. A decision model doesn't write anything. You give it an input and a list of possible answers, and it returns a probability score for each option as JSON. Microsoft's model catalogue describes it as a model that "returns calibrated probability scores for fixed answer options". Calibrated means a score of 0.8 should be right about 80% of the time, so you can set a threshold for when a person needs to check.
Example uses include routing a support ticket to the right team, checking whether a document meets a rule, or flagging which requests need a human.
Quick check · Side by side
Soft opening or grand opening: which suits the job?
Tap what matters to you (pick one or more) and see which option fits.
Preview or beta
- Yes, often months before launch
- Feedback is the point of a preview
- Usually meant for testing only
- Usually none
- Can change or be removed at any time
Generally available (GA)
- Arrives later, once the product is finished
- Main design is already settled
- Declared ready for production use
- Included where the vendor offers one
- Stable, documented and supported
Pick what matters to you and the better fit lights up.
From the explainer What is general availability (GA)? Explained with a restaurant opening
What's new in 1.1
- Speed: Microsoft says it is more than twice as fast as its first decision model, at the same cost.
- Images: it now accepts photos and screenshots as well as text, up to 8 images per call according to the Foundry catalogue page.
- Quality: Microsoft reports a score of 68 on the Artificial Analysis Decision Index, which rates quality, calibration and speed, six points higher than its first model.
- Uptake: Microsoft says more than 4,000 customers have used it since the public preview opened last month.
The catalogue says the model is built on Google's open-weight Gemma 4 26B-A4B and then further trained by Microsoft. Inputs can be up to about 32,000 tokens (a token is roughly three-quarters of a word).
What it costs
Microsoft's announcement gives no per-token price, only that the new version costs the same as the first one. The Foundry page offers pay-as-you-go or reserved capacity and points to Azure's model pricing page. Check the current price in Foundry, OpenRouter or Vercel before you plan a budget.
Who should care
Teams that already run many small yes/no or pick-one checks through a general chat model. A dedicated decision model can be faster and cheaper for that work, and the probability scores make it easier to send only uncertain cases to a person. Google launched a similar product, the Gemini Decisions API, last week, so there are now two big-vendor options to compare.
Limits and caveats
- It can't write text or answer open questions. The catalogue lists text generation and open-ended Q&A as out of scope.
- Microsoft says it should not be the sole automated decision-maker for credit, employment, housing, insurance, education, healthcare or legal matters. Keep a person in the loop there.
- The speed and figures are Microsoft's own claims. Test on your own data.
- No regional availability details are given in the announcement.
What to do this week
- List the classification or routing steps in your workflows that currently use a full chat model.
- Take a few hundred past cases with known answers and run them through Decision-1.1 to measure accuracy and cost.
- Pick a confidence threshold below which a person reviews the case, and log every decision.
- If you're also looking at Google's option, test both on the same set.
Sources
You’ve read the news. Now understand it.
- Learn · 5 min · Side by sideWhat is general availability (GA)? Explained with a restaurant openingGeneral availability (GA) is when software is open to all customers and declared ready for real work, with support that preview and beta versions lack. Explained with a restaurant's soft opening and grand opening.Start in 5 seconds →
- Learn · 5 min · Everyday ↔ realWhat is an API? Explained with a waiter in a restaurantAn API is a set of rules that lets one program ask another for data or a task and get an answer back. Explained with a waiter carrying orders between you and the kitchen.Start in 5 seconds →
- Learn · 5 min · Myth or factWhat are AI benchmarks? Explained with a school examAn AI benchmark is a standard test: the same questions and the same marking for every model, so scores can be compared. Explained with a school exam that every student sits.Start in 5 seconds →


