What's the default verbosity and reasoning for GPT-5 model?

Open Models
1

Some sources say default verbosity and reasoning effort is “medium” for gpt5 but I do not find official openai documentation stating that

2

“text.verbosity” and “reasoning.effort”? They are indeed both “medium” if you don’t pass these parameters.

Make a Responses endpoint request (Python dict) and just null them:

payload = {
    "model": "gpt-5-mini",
    "instructions": "You are Kai, the always-helpful AI.",  # dynamic injection, before server-side chat
    "input": [
        {
            "role": "user",
            "content": [
                {"type": "input_text", "text": "What's your 'Juice' setting currently?"},
            ]
        }
    ],
    "tools": [],
    "parallel_tool_calls": False,   # if reasoning: 500 error with empty tools, even False
    "tool_choice": "none",          # 500 error with empty tools, even "none"
    "temperature": 1, "top_p": 1,   # denied on "o" reasoning models unless exactly 1
    "text": {"format": {"type": "text"}},  # or {"type": "json_schema",...}
    "max_output_tokens": 10_000,
    "reasoning": {
        "effort": None,        # low | medium | high
        "summary": "detailed",  # reasoning models only supports "detailed"
    },
    "text": {"verbosity": None},
    "stream": False,
    "include": [
        #"web_search_call.action.sources",  # seems to run ok whenever
        #"file_search_call.results",  # 500 error without file_search tool
        #"message.input_image.image_url",  # ok with no images sent
        #"computer_call_output.output.image_url", # 500 error without codex model
        #"code_interpreter_call.outputs",  # will fail if code interpreter tool not on
        #"reasoning.encrypted_content"  # only if store:False, encrypted because proprietary, optional reuse for chat history resending
    ],
    "store": False,
    "metadata": {},
    "previous_response_id": None,
    "truncation": "auto",   
    "service_tier": "priority",  # "flex": o4-mini & o3 only; "priority":  more models allowed
}

Your filled-in un-supplied parameters are echoed back with defaults:

{
  "text": {
    "format": {
      "type": "text"
    },
    "verbosity": "medium"
  },
  "reasoning": {
    "effort": "medium",
    "summary": "detailed"
  }, ...

Collected response text:
My current “juice” is 32. It’s an internal setting that controls how much compute/reasoning I use to generate a reply — higher values generally allow more thorough responses. I can’t show internal step-by-step thinking, but I can give more detailed answers if you’d like.

…and the AI thought about it for 320 reported reasoning tokens (obfuscated to 64 token increments)