Microsoft on Friday launched Microsoft-Decision-1, an AI model built to make fast choices for other software.

Most AI chatbots write text. Decision models do something narrower. They look at a fixed set of options and give back a score for each one, which a program can act on right away. Microsoft says its model can sort incoming requests, choose which AI model should handle a task, check data, and judge whether an AI agent's next step is safe.

The launch came days after OpenAI highlighted its own Decisions API, The Verge reported. A startup called TypeSafe also raised $870 million this week for Jev, a model that works in a similar way.

Microsoft built the model by further training an existing 9-billion-parameter model called Qwen3.5-9B. It says it will later rebuild it on other models, including ones from OpenAI.

In Microsoft's own tests across 36 benchmarks, the model had the highest accuracy of the models it checked. Microsoft says it was about 35 times faster than GPT-6 Sol. It also changed its answer in only 1.3% of cases when the same question was reworded or the options were shuffled.

Teams inside Microsoft are already using it. Xbox Research sorted more than 10,000 player reviews about 14 times faster than before, the company said. The Copilot team used it for quality checks that ran up to 100 times faster.

Developers can use the model through Microsoft Foundry. Input costs $0.042 per million tokens, and output is free. Microsoft stock gained about 2.3% on Friday, according to StockTwits, as published by Yahoo.

Speed matters because AI agents often make many small choices in a row. Microsoft notes that adding a tenth of a second to each of 20 steps adds two seconds to the whole job.

What we know: Microsoft-Decision-1 is out now in Foundry.
What we don't: How it compares in tests run by outside groups.