Agent Max Tokens Invalid Input Range

[leoglaz]

Severity

P3 - Medium (Minor functionality affected)

Describe the Bug with repro steps

Excepted Behavior:

• MAX Tokens will be reflective of the MAX Tokens limits for the model selected in the agent configuration. For gpt-4o - 16k and for gpt-4.1 32k
  Actual Behaviour:
• UI is limiting the MAX Tokens to 8192

Steps to Reproduce:

• in the azure portal create an autonomous agent
• configure the agent
  • select GPT-4o or GPT-4.1
  • Advanced Parameters -> MAX TOKENS (max limit is always 8192 regardless of the model selected)

What type of Logic App Is this happening in?

Consumption (Portal)

Are you experiencing a regression?

No response

Which operating system are you using?

Windows

Did you refer to the TSG before filing this issue? https://aka.ms/lauxtsg

Yes

Workflow JSON

{
    "definition": {
        "$schema": "https://schema.management.azure.com/providers/Microsoft.Logic/schemas/2016-06-01/workflowdefinition.json#",
        "contentVersion": "1.0.0.0",
        "actions": {
            "Agent": {
                "type": "Agent",
                "inputs": {
                    "parameters": {
                        "deploymentId": "gpt-4.1",
                        "messages": [
                            {
                                "role": "system",
                                "content": "test"
                            }
                        ],
                        "agentModelType": "AzureOpenAI",
                        "agentModelSettings": {
                            "agentHistoryReductionSettings": {
                                "agentHistoryReductionType": "maximumTokenCountReduction",
                                "maximumTokenCount": 128000
                            },
                            "deploymentModelProperties": {
                                "name": "gpt-4.1",
                                "format": "OpenAI",
                                "version": "2025-04-14"
                            }
                        }
                    },
                    "modelConfigurations": {
                        "model1": {
                            "referenceName": "agent"
                        }
                    }
                },
                "tools": {},
                "runAfter": {},
                "limit": {
                    "count": 100
                }
            }
        },
        "outputs": {},
        "triggers": {
            "When_an_HTTP_request_is_received": {
                "type": "Request",
                "kind": "Http"
            }
        }
    },
    "kind": "Stateful"
}

Screenshots or Videos

No response

Environment

• Microsoft Edge for Business Version 146.0.3856.84

Additional context

when using the agentic activity we are encountering token limits; this is resulting in us having to use the Azure Open AI - Get Chat Completions call where we are able to set the MAX Tokens value to the desired upper limit

[github-actions]

AI Triage Analysis

Component: Agentic (Agent Loop) | Priority: P3 - Medium | SKU: Consumption (Portal), likely Standard too

Relevant Code

• libs/logic-apps-shared/src/designer-client-services/lib/consumption/manifests/agentloop.ts:L90-L96 — maxTokens property hardcoded with maximum: 8192 and description "value should be between 1 and 8192"
• libs/logic-apps-shared/src/designer-client-services/lib/standard/manifest/agentloop.ts:L215-L221 — Same hardcoded maximum: 8192 in the Standard agentloop manifest
• libs/logic-apps-shared/src/utils/src/lib/models/agent.ts — AGENT_MODEL_CONFIG defines model metadata (name, version, format) but no per-model token limits

Probable Root Cause

Both agentloop operation manifests define the maxTokens input field with a static maximum: 8192 constraint, regardless of the model selected via deploymentId. The AGENT_MODEL_CONFIG map doesn't include a maxTokens or contextWindow property per model, so there's no mechanism to dynamically update the parameter's max constraint when the user picks gpt-4.1 (32k) or gpt-4o (16k). The UI schema validation simply enforces the hardcoded manifest limit.

Suggested Investigation

1. Add per-model token limit info to AGENT_MODEL_CONFIG in libs/logic-apps-shared/src/utils/src/lib/models/agent.ts (e.g., a maxOutputTokens or contextWindow field per model entry).
2. Update the maxTokens manifest field in both consumption/manifests/agentloop.ts and standard/manifest/agentloop.ts to use an x-ms-input-dependencies dynamic max constraint driven by the selected deploymentId, or raise the static maximum to match the highest supported model limit.
3. Also update the description string in both manifests (currently says "between 1 and 8192") to reflect the actual model-dependent range.

Similar Issues

No exact duplicates found.

---

Automated analysis by issue triage agent. May contain inaccuracies — verify before acting.
Calibration: reviewed 10 past triages, 0 corrections applied.

Generated by Issue Triage Agent for Azure Logic Apps UX for issue #9025 · ◷