Hi, is there any way put a limit and do a stop based on cost in Azure AI Foundry GPT model deployed ?

Hello @Sen0299 ,

Welcome to Microsoft Q&A .Thank you for reaching out to us.

Azure AI Foundry / Azure OpenAI does not currently provide a native real-time dollar-based hard stop mechanism that automatically blocks inference requests when a spending threshold is reached. Azure Cost Management Budgets can monitor spend and generate alerts, but budgets do not automatically stop deployments, disable endpoints, or block model traffic. Cost-based enforcement requires additional automation.

The most practical and commonly adopted production pattern combines monitoring, automation and real-time usage controls.

Layer 1 – Azure Cost Management Budgets -Monitoring

Create budgets at the appropriate scope:

• Subscription
• Resource Group
• Specific Resource

Configure spending thresholds such as:

• 50%
• 80%
• 100%

Budgets provide financial visibility and generate notifications when thresholds are reached. However, budget alerts are informational and do not enforce shutdown actions

Layer 2 – Azure Monitor Action Groups - Automation

udget alerts can trigger Azure Monitor Action Groups, which can launch automated workflows through:

• Azure Logic Apps
• Azure Functions
• Azure Automation Runbooks

Depending on operational requirements, automation can be configured to:

• Restrict endpoint access.
• Stop application routing to the deployment.
• Apply network controls.
• Temporarily prevent new inference traffic.
• Execute other governance actions appropriate for the environment.

This creates an effective Budget > Alert > Action workflow.

Layer 3 – Azure API Management -Real-Time Control

For immediate protection against unexpected spikes or runaway consumption, Azure API Management (AI Gateway pattern) can be placed in front of the deployment.

This enables:

• Request throttling
• Token-based quotas
• Immediate HTTP 429 enforcement
• Real-time usage controls

This mechanism controls usage (tokens and requests), rather than direct currency spend, but it provides the closest form of real-time enforcement available

Please note that -

1. TPM/RPM quotas are throughput controls rather than spending controls.
2. Budgets are monitoring and alerting mechanisms, and are not enforcement mechanisms.
3. Budget evaluations rely on billing and usage ingestion and are therefore not instantaneous.
4. A small overspend beyond a configured threshold may occur before automation executes.
5. The most effective governance model combines Budgets + Action Groups + API Management

Regarding , GPT‑4.1‑mini 1 million token context window

The GPT‑4.1‑mini model supports a 1 million token context window as part of its built-in model capability. Current public documentation does not indicate a separate approval requirement specifically for using the supported 1 million token context length.

It is helpful to distinguish between context window size and quota allocation, as they are independent controls.

• Context Window
  • Defines the maximum amount of information that can be processed in a single request.
  • GPT‑4.1‑mini supports up to 1 million tokens.
  TPM/RPM Quota
  • Controls throughput and request volume.
  • Managed independently from context size.
  • Additional TPM quota may be needed for higher-volume workloads, but quota allocation does not change the model’s supported context window.

The following references might be helpful , please check them out

• Tutorial - Create and manage budgets - Microsoft Cost Management | Microsoft Learn
• Create and manage action groups in Azure Monitor - Azure Monitor | Microsoft Learn
• Manage Azure OpenAI in Microsoft Foundry Models quota (classic) - Microsoft Foundry (classic) portal | Microsoft Learn
• Azure OpenAI in Microsoft Foundry Models Quotas and Limits - Microsoft Foundry | Microsoft Learn
• Azure API Management policy reference | Microsoft Learn

Please let us know if the response was helpful

Thank you