Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- The model scores fixed options and returns structured probabilities rather than free-form explanations; the current Foundry catalog lists it as Preview.
- Microsoft Learn documents noul, choice, and score question types. The catalog lists text input, JSON output, and a 32,768-token context window.
- Deploy in Foundry and call the systemone endpoint using your deployment name as model; confirm permissions, project access, and Region in your account.
- Microsoft's October 9 launch post lists $0.042 per million input tokens and free output tokens as a reference; recheck current account pricing and other Azure charges.
Start with the status: usable in Foundry, but still marked Preview
Microsoft-Decision-1 is a decision-scoring model for fixed answer options. Give it a state and a bounded question, and it returns a probability score for each option instead of free-form text or a written rationale. Its uses include classification, ranking, prioritization, and workflow routing.
Microsoft's October 9, 2026 launch post says developers can get started in Microsoft Foundry that day. The current Foundry catalog still lists its Lifecycle as Preview. Deployment depends on your subscription, project, permissions, Region and quota.
Shape the state and question as a bounded scoring task
Microsoft Learn documents three question types: noul for yes-or-no decisions, choice for selecting from fixed categories, and score for ordered levels. The request state can be text or JSON. Answers are returned by question name as probabilities or scores for the application to handle with its own thresholds.
The Foundry catalog lists text input, JSON output, and a 32,768-token context window. This is not a chat model that writes explanations, and it does not accept images, audio, or video as direct inputs. When evidence is insufficient, include a cannot tell option so the workflow has a way to abstain.
Call the deployed model with its resource endpoint and name
Deployment requires an Azure subscription with a valid payment method, access to Microsoft Foundry and a project, and permission to create or manage deployments. Cognitive Services Contributor can deploy models. Microsoft Learn recommends Microsoft Entra ID and also supports a resource API key; confirm the role and authentication method for the target resource.
Copy the resource endpoint and deployment name from deployment details, then send a request to <resource-endpoint>/providers/microsoft/v1/systemone. In the request, model must contain your deployment name, not the underlying Microsoft-Decision-1 model name.
Choose a deployment type and Region for your data requirements
Microsoft Learn lists GlobalStandard and DataZoneStandard in selected Regions. GlobalStandard can process inference in any supported Azure Region where the model is deployed, so latency may be higher or more variable. When available, DataZoneStandard keeps inference processing within the selected data zone.
Regional availability varies by deployment type and Azure Region; a published matrix does not prove that a particular subscription is eligible. Check the current model availability table and target Foundry project for deployment options, capacity, and processing location before choosing based on residency requirements.
Treat the October price as a reference, not a current quote
Microsoft's October 9 launch post lists input tokens at $0.042 per million and output tokens as free. Those are the figures stated in that post for Microsoft-Decision-1; they do not make an entire Foundry workflow free or establish the bill for every subscription, agreement, or Region.
The current Foundry catalog links to pricing for the account and deployment context but does not show the same numeric rate in its public text. Because the model lifecycle remains Preview, recheck portal pricing, metering, and quota before rollout. Azure services or other workflow steps can add separate costs.
Use scores to inform a workflow, not grant business authority
Start with redacted, labeled examples to check category definitions, probability thresholds, and the path for uncertain cases. A probability is not business accuracy, proof of a fact, or permission to execute an action. Keep the appropriate review and business rules for consequential decisions such as refunds, account bans, or hiring.
OpenAI's Decisions API is a separate product and endpoint; its request fields, models and billing should not be copied into Foundry. The OpenAI Decisions API integration guide under Further reading explains that separate interface. Verify each product against its own official documentation and project access.
Sources
Frequently Asked Questions
Is Microsoft-Decision-1 generally available?
The current Microsoft Foundry model catalog lists its Lifecycle as Preview. The launch post says developers can get started, but that does not establish that the lifecycle has changed to GA. Check the current catalog and your account before deployment.
Does it return a written explanation for each score?
It does not generate free-form explanations. It returns structured scores or probabilities for fixed options. Your application must provide evidence display, thresholds, and review paths.
Does free output mean that a call has no cost?
No. Microsoft's October 9 launch post states $0.042 per million input tokens and free output tokens. Confirm current Foundry deployment pricing, quota, and separate Azure service costs.
Does GlobalStandard guarantee processing in one chosen Region?
No. Microsoft Learn says GlobalStandard can process inference in any supported Azure Region. DataZoneStandard keeps processing within the corresponding data zone when available. Check the current matrix and project configuration.