GPT-5 mini returns incomplete response when using web_search_preview due to max_output_tokens limit

API
,
1

I’m using GPT-5 mini with the Responses API and web_search_preview tool, but consistently
getting incomplete responses even with max_output_tokens set to 8000.

Problem

The model uses most tokens for reasoning and web searches, leaving no room for the actual JSON
output.

Code

const response = await openai.responses.create({
  model: 'gpt-5-mini',
  max_output_tokens: 8000,
  tools: [{
    type: 'web_search_preview',
    search_context_size: 'high'
  }],
  input: "Search for events in Kagoshima...",
  text: {
    format: {
      type: 'json_schema',
      name: 'EventList',
      schema: { /* ... */ },
      strict: true
    }
  }
})

Response

{
  "status": "incomplete",
  "incomplete_details": { "reason": "max_output_tokens" },
  "output": [
    { "type": "reasoning" },
    // 15-20 web_search_call items
    // No message with JSON output
  ],
  "usage": {
    "output_tokens": 8000,
    "output_tokens_details": {
      "reasoning_tokens": 7500+
    }
  }
}

Question

How can I get complete JSON output when using web_search_preview with GPT-5 mini? Is there a
way to limit reasoning/search tokens to preserve space for the actual output?

Environment: OpenAI SDK 4.77.0, Node.js