Wire requests

One normalized request, rendered into both OpenAI dialects. They are implemented as separate builders sharing no object literals, because conflating them is the single largest source of 400s — and because a framework that claims to “craft requests correctly” should let you check.

streaming

The baseline. Note the tool shape: Responses puts name/parameters at the top level, Chat Completions nests them under `function`. Sending one to the other is a guaranteed 400.

Input — NormalizedRequest

{
  "model": "gpt-5.2",
  "system": "You are a research assistant.",
  "messages": [
    {
      "id": "user_1",
      "role": "user",
      "parts": [
        {
          "type": "text",
          "text": "Which vendor should we pick?"
        }
      ]
    }
  ],
  "tools": [
    {
      "name": "compareCompetitors",
      "description": "Positioning matrix.",
      "parameters": {
        "type": "object",
        "properties": {
          "against": {
            "type": "array",
            "items": {
              "type": "string"
            }
          }
        }
      }
    }
  ],
  "maxOutputTokens": 2000
}

POST /v1/responses

{
  "model": "gpt-5.2",
  "input": [
    {
      "type": "message",
      "role": "user",
      "content": [
        {
          "type": "input_text",
          "text": "Which vendor should we pick?"
        }
      ]
    }
  ],
  "instructions": "You are a research assistant.",
  "tools": [
    {
      "type": "function",
      "name": "compareCompetitors",
      "description": "Positioning matrix.",
      "parameters": {
        "type": "object",
        "properties": {
          "against": {
            "type": "array",
            "items": {
              "type": "string"
            }
          }
        }
      }
    }
  ],
  "tool_choice": "auto",
  "parallel_tool_calls": true,
  "max_output_tokens": 2000
}

POST /v1/chat/completions

{
  "model": "gpt-5.2",
  "messages": [
    {
      "role": "system",
      "content": "You are a research assistant."
    },
    {
      "role": "user",
      "content": "Which vendor should we pick?"
    }
  ],
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "compareCompetitors",
        "description": "Positioning matrix.",
        "parameters": {
          "type": "object",
          "properties": {
            "against": {
              "type": "array",
              "items": {
                "type": "string"
              }
            }
          }
        }
      }
    }
  ],
  "parallel_tool_calls": true,
  "tool_choice": "auto",
  "max_completion_tokens": 2000
}

Where they disagree

	Responses	Chat Completions
history field	`input[]`	`messages[]`
system	`instructions (top level)`	`a message with role system`
user text part	`input_text`	`text`
image part	`input_image`	`image_url`
file part	`input_file`	`file`
assistant text	`output_text`	`content: string`
tool declaration	`flat { type, name, parameters }`	`nested { type, function: {…} }`
model tool call	`input item function_call`	`assistant.tool_calls[]`
tool output	`input item function_call_output`	`message role:"tool" + tool_call_id`
output cap	`max_output_tokens`	`max_completion_tokens`
JSON schema	`text.format`	`response_format.json_schema`
reasoning	`reasoning.effort`	`reasoning_effort`
streamed usage	`always included`	`needs stream_options.include_usage`
async	`background: true + polling`	`— none`
server history	`previous_response_id`	`— none`