When routing requests through OpenClaw’s Azure OpenAI Responses adapter to GPT-5.2-Codex or GPT-5.3-Chat deployments, the gateway returns HTTP 400 errors. The failure manifests in two distinct patterns:

Pattern A: Immediate 400 on First Request

warn agent/embedded {"event":"embedded_run_agent_end","isError":true,"error":"400 Item 'rs_07f091ad1d9adbcb0069d7059e74a08190a5fd477877af8e27' of type 'reasoning' was provided without its required following item.","failoverReason":"format","model":"gpt-5.3-chat","provider":"AzureOpenAI-Three"}

Pattern B: 400 on Subsequent Turns

The first user turn succeeds, but subsequent conversational turns fail with the same reasoning item error after the assistant attempts to continue the conversation.

Error Signature

The error hash sha256:ce60f0254cd4 is consistent across multiple failures, indicating a systematic payload construction issue rather than transient network errors.

Affected Configuration

yaml

Provider configuration triggering the error

“AzureOpenAI-Three”: { “baseUrl”: “https://dy-aoai.openai.azure.com/openai/v1", “api”: “openai-responses”, # or “azure-openai-responses” “auth”: “api-key”, “models”: [{ “id”: “gpt-5.3-chat”, “reasoning”: false, # Explicitly disabled “contextWindow”: 1048576 }] }

Control UI Behavior

Users observe the error through the OpenClaw Control UI when:

1. Selecting gpt-5.2-codex or gpt-5.3-chat as the active model
2. Submitting any prompt (even simple test queries like “Hello, respond with ‘OK’”)
3. The WebSocket disconnects with code 1001 during extended conversations

The HTTP 400 error originates from a mismatch between the request payload structure generated by OpenClaw’s azure-openai-responses adapter and the strict validation enforced by Azure OpenAI’s Responses API endpoint.

Technical Analysis of the Failure Sequence

The Azure OpenAI Responses API (v1 API format) enforces a structural constraint on reasoning items:

1. Reasoning Item Requirement: When a reasoning item appears in the output array, Azure OpenAI requires a corresponding text or output_text item immediately following it.

2. Adapter Payload Generation: The azure-openai-responses adapter constructs request payloads that include reasoning items in the output array, even when the model configuration specifies "reasoning": false.

3. Missing Following Item: The generated payload contains: json { “output”: [ { “type”: “reasoning”, “id”: “rs_07f091ad1d9adbcb0069d7059e74a08190a5fd477877af8e27”, “summary”: [] } // Missing required “text” or “output_text” item here ] }

4. Azure Validation Rejection: Azure OpenAI’s backend rejects the request with:

   400 Item ‘rs_07f091ad1d9adbcb0069d7059e74a08190a5fd477877af8e27’ of type ‘reasoning’ was provided without its required following item.

Root Cause Locations

Architectural Inconsistency

The adapter’s payload construction logic does not respect the model’s reasoning flag. The code path assumes all GPT-5 variants with extended context windows should include reasoning blocks, regardless of the explicit configuration:

typescript // Hypothetical problematic code path in adapter function buildResponsePayload(model, messages, options) { // ISSUE: No check for model.reasoning === false if (model.contextWindow > 128000) { outputItems.push({ type: “reasoning”, summary: [] }); } // Reasoning item added without required following text item }

Affected Request Flow

User Prompt → Control UI → OpenClaw Gateway → azure-openai-responses adapter → Payload construction (reasoning item without following item) → Azure OpenAI endpoint → 400 Validation Error

Option 1: Disable Reasoning at Adapter Level (Recommended)

Modify the provider configuration to explicitly disable reasoning processing:

Before: json { “AzureOpenAI-Three”: { “baseUrl”: “https://dy-aoai.openai.azure.com/openai/v1", “api”: “azure-openai-responses”, “models”: [{ “id”: “gpt-5.3-chat”, “reasoning”: false }] } }

After: json { “AzureOpenAI-Three”: { “baseUrl”: “https://dy-aoai.openai.azure.com/openai/v1", “api”: “azure-openai-responses”, “compat”: { “reasoningEnabled”: false }, “models”: [{ “id”: “gpt-5.3-chat”, “reasoning”: false, “reasoningEffort”: null }] } }

Option 2: Use Standard OpenAI-Responses Adapter

If the azure-openai-responses adapter continues to fail, configure the provider to use the standard openai-responses adapter with explicit output formatting:

Configuration: json { “AzureOpenAI-Three”: { “baseUrl”: “https://dy-aoai.openai.azure.com/openai/v1", “api”: “openai-responses”, “auth”: “api-key”, “headers”: { “api-key”: “YOUR-API-KEY”, “Azure-Extensions-Version”: “2024-11-01” }, “models”: [{ “id”: “gpt-5.3-chat”, “reasoning”: false, “outputFormat”: “text” }], “requestOptions”: { “stripReasoningItems”: true } } }

Option 3: Patch Adapter Configuration via Environment Variables

For deployments without config file access:

bash

Set environment variable before starting OpenClaw

export OPENCLAW_AZURE_RESPONSES_STRIP_REASONING=true export OPENCLAW_AZURE_RESPONSES_OUTPUT_FORMAT=text

Restart OpenClaw Gateway

openclaw gateway restart

Option 4: Direct Provider Config Update

Edit ~/.openclaw/config.yaml or the active configuration file:

yaml providers: AzureOpenAI-Three: type: azure-openai-responses baseUrl: “https://dy-aoai.openai.azure.com/openai/v1" apiKey: “${AZURE_OPENAI_API_KEY}”

modelDefaults:
  reasoning: false
  reasoningEffort: null

adapterOptions:
  outputFormat: "text"
  requireFollowingItem: true
  stripReasoningItems: true

models:
  - id: "gpt-5.3-chat"
    name: "GPT-5.3-Chat (Azure dy-aoai)"
    reasoning: false
    maxTokens: 131072
  
  - id: "gpt-5.2-codex"
    name: "GPT-5.2-Codex (Azure dy-aoai)"
    reasoning: false
    maxTokens: 131072

Option 5: Runtime Fix via OpenClaw CLI

bash

Update provider configuration via CLI

openclaw config set-provider AzureOpenAI-Three
–adapter azure-openai-responses
–set reasoning=false
–set adapterOptions.stripReasoningItems=true

Verify the update

openclaw config get-provider AzureOpenAI-Three

Restart gateway to apply changes

openclaw gateway restart

After applying the fix, verify the resolution using the following validation steps:

Step 1: Restart Gateway and Check Startup Logs

bash

Restart the OpenClaw Gateway

openclaw gateway restart

Monitor startup logs for successful initialization

tail -f ~/.openclaw/logs/gateway.log | grep -E “(startup|adapter|AzureOpenAI)”

Expected Output:

info gateway/startup {“message”:“Gateway started”,“port”:18792} info adapter/azure-openai-responses {“provider”:“AzureOpenAI-Three”,“status”:“initialized”,“outputFormat”:“text”,“stripReasoningItems”:true}

Step 2: Test via OpenClaw Control UI

1. Open OpenClaw Control UI in browser
2. Select gpt-5.3-chat or gpt-5.2-codex from the model dropdown
3. Send test prompt: "Hello, respond with just 'OK'"
4. Confirm successful response without 400 error

Step 3: Verify via API Call

bash curl -X POST http://localhost:18792/v1/chat/completions
-H “Content-Type: application/json”
-H “Authorization: Bearer ${OPENCLAW_API_KEY}”
-d ‘{ “model”: “gpt-5.3-chat”, “messages”: [{“role”: “user”, “content”: “Respond with OK”}], “provider”: “AzureOpenAI-Three” }’

Expected Response (200 OK): json { “id”: “chatcmpl-…”, “object”: “chat.completion”, “model”: “gpt-5.3-chat”, “choices”: [{ “message”: {“role”: “assistant”, “content”: “OK”}, “finish_reason”: “stop” }] }

Step 4: Verify Request Payload Structure

Enable debug logging to inspect the actual payload sent to Azure:

bash

Enable debug logging

export OPENCLAW_LOG_LEVEL=debug

Run a test request

openclaw chat “Hello” –model gpt-5.3-chat –provider AzureOpenAI-Three

Expected Debug Output (payload sent to Azure): json { “model”: “gpt-5.3-chat”, “input”: { “messages”: […] }, “output”: [ // No reasoning items present when reasoning: false ] }

Step 5: Verify No Reasoning Items in Logs

bash

Check logs for reasoning item errors (should be absent)

grep -i “reasoning.*required following item” ~/.openclaw/logs/gateway.log

Expected: no output (no matching lines)

Exit Code Check: bash

Verify no 400 errors in recent logs

grep -c “400.*reasoning” ~/.openclaw/logs/gateway.log

Expected: 0

• HTTP 400: "Invalid request format" — General Azure OpenAI request validation failure. May indicate payload structure issues beyond reasoning items.
• HTTP 400: "Unsupported model" — Model ID not recognized by the Azure OpenAI deployment. Verify deployment name matches configuration.
• HTTP 401: "Authentication failed" — Invalid or expired API key. Ensure `api-key` header is properly set for Azure endpoints.
• HTTP 422: "Content filter triggered" — Azure content moderation blocked the request. Check prompt content and Azure content filters.
• Error: "Item of type 'reasoning' was provided without its required following item" — The specific error documented in this guide. Indicates malformed reasoning block in payload.
• Error: "Token limit exceeded" — Request exceeds model's context window or maxTokens. Reduce prompt size or adjust `maxTokens` in config.
• WebSocket 1001** — Gateway disconnected during conversation. May result from unhandled 400 errors propagating to Control UI.
• Error: "ENOENT: no such file or directory"** — Session memory file access failure. Unrelated to the reasoning item bug but may appear in same logs.

Historical References:

• GitHub Issue #1847: Azure OpenAI Responses adapter payload validation issues
• GitHub Issue #1923: Reasoning items not properly stripped from multi-turn conversations
• GitHub Discussion #456: GPT-5 model compatibility with OpenClaw adapters
• Pull Request #2101: Fix reasoning item following-item requirement validation