Jev Daily

Microsoft-Decision-1.1 goes live on OpenRouter

A second vendor is now selling the one call decision shape, and it shipped without a single published number.

OpenRouter says Microsoft-Decision-1.1 is generally available, one of three Microsoft AI models announced at Ignite that went live on the router the same day. The other two are image models, MAI-Image-2.7, which generates and edits at up to 2K resolution with transparent backgrounds, and MAI-Image-2.7 Flash, described as the faster, lower cost version for high volume work.

OpenRouter
@OpenRouter
X
Instead of generating text, it reads your input and returns a calibrated probability for each answer option, so you get a clear decision in a single fast call.
Nov 17, 2026 · View on X

Decision-1.1 is the one that matters here. By OpenRouter's description it does not generate text at all. It reads the input and returns a calibrated probability for each answer option, in a single call. That is the same shape Jev takes, a model that returns a distribution over the options you hand it rather than a string you then have to parse and trust.

What was not in the post

No accuracy figure. No latency figure. No price. No eval, no task list, and no comparison against anything else serving the same shape. The announcement is a product listing with a description of the interface and three links to try the models.

So the calibration claim is currently the vendor's claim, relayed by the router. Calibrated has a specific meaning worth holding people to, that when the model says 0.7 it is right about seventy percent of the time across a large bucket of those predictions. Nobody in our sources has run Decision-1.1 on their own data yet, which means nobody has checked whether the buckets hold on anything other than Microsoft's internal evaluation.

For people building on it

If you are already routing a classification path through Jev, this is a listing to note and not yet a migration to plan. The useful comparison is not the marketing shape, which is now identical across vendors, it is accuracy on your questions at your price, and that number does not exist in public. Google's entry in the same category arrived with at least a price attached, earlier at $0.10 per million input tokens. Microsoft's arrived with none.

The first person to run the same question set through both and publish the buckets will tell you more than the launch post did.

Know what Jev is good at before you bet on it

One email a day. Under three minutes. What builders found out about Jev yesterday.

Get the next one by email

Jev, read daily so you do not have to. The builds, the benchmarks, the criteria that worked and the cases where it lost, from the people shipping on TypeSafe AI's System One model.