Thesis: The useful change is a model interface that scores a fixed choice set for downstream software. Observed facts: The report says the model returns calibrated probabilities for fixed options and supports yes/no, multiple-choice, rating, and rubric-based scoring. It also says Microsoft post-trained it on Qwen3.5-9B and made it available in Microsoft Foundry. Mechanism: A caller supplies bounded alternatives; the response assigns scores that a workflow can use to route, defer, retry, or escalate. Why now: Agent systems need outputs that can enter control flow directly, and Microsoft is presenting Decision-1 as a low-cost scoring component for that role.
Implications: First, structured scoring could shift some routing and evaluation steps from general-purpose generation to a narrower model call. Second, probability outputs make calibration and the cost of wrong decisions operational concerns, not merely benchmark details. The report describes internal use in Xbox feedback labeling and gives pricing from $0.042 per million input tokens, with output tokens free. Evidence boundary: Performance, consistency, and internal workflow outcomes in the supplied report are attributed to Microsoft; this source does not provide independent replication or enough detail to assess benchmark comparability. Watch condition: Independent evaluations should test calibration, robustness to changed options, and end-to-end error rates in real workflows.
The release starts a distinct analytical record because no prior New Runtime record is supplied for this specific model or its fixed-option scoring interface. The source says OpenRouter support is planned, but does not establish its timing or availability. Source: https://www.testingcatalog.com/microsoft-launches-decision-1-model-in-foundry