Uploaded October 2026 | Updated November 2026, 10 hours ago
Microsoft just joined a new AI trend: small, fast decisions.
We've seen Jev and OpenAI's Decisions API. Now there's Microsoft-Decision-1.
The idea is practical: give the model context and possible answers, get scores back, then let your software decide what happens next. Is this spam? Which team should handle a support request? Should an AI agent continue or ask for help?
Microsoft lists the price at 4.2 cents per million input tokens, with no output charge. What interests me is having more places in an app where a quick check is practical.
Microsoft's model is text-only. OpenAI's Decisions API also accepts images, which opens up things like checking product photos for damage or organizing screenshots.
As a developer, I'm excited about what we can build. I also think cheaper analysis could mean more of our posts, uploads, and interactions being classified. That could make software more useful, and those predictions can still be wrong.
When a prediction affects a person, how should we catch and correct the mistake?
#AI #SoftwareDevelopment
Microsoft just joined a new AI trend: small, fast decisions.
We've seen Jev and OpenAI's Decisions API. Now there's Microsoft-Decision-1.
The idea is practical: give the model context and possible answers, get scores back, then let your software decide what happens next. Is this spam? Which team should handle a support request? Should an AI agent continue or ask for help?
Microsoft lists the price at 4.2 cents per million input tokens, with no output charge. What interests me is having more places in an app where a quick check is practical.
Microsoft's model is text-only. OpenAI's Decisions API also accepts images, which opens up things like checking product photos for damage or organizing screenshots.
As a developer, I'm excited about what we can build. I also think cheaper analysis could mean more of our posts, uploads, and interactions being classified. That could make software more useful, and those predictions can still be wrong.
When a prediction affects a person, how should we catch and correct the mistake?
#AI #SoftwareDevelopment










