Response Data Handling

This page explains how to correctly parse the response data returned by 4ALL API, and applies to all models called via Chat Completions.

Non-streaming Response Structure

{
  "id": "chatcmpl-xxx",
  "object": "chat.completion",
  "model": "gpt-5.5",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "……" },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 46,
    "total_tokens": 58
  }
}

finish_reason Handling

Value	Meaning	Recommended Handling
stop	Ended normally	Use directly
length	Truncated after reaching the max_tokens limit	Increase max_tokens or continue writing in segments
tool_calls	The model requested a function call	Return the result after executing the function, see FC Function Calling
content_filter	Content was blocked by upstream safety policies	Adjust the prompt

usage Fields

• usage is the basis for reconciliation: billing is charged separately by prompt_tokens / completion_tokens;
• The final data chunk of a streaming call also carries usage (see Streaming Output);
• When a prompt cache is hit, prompt_tokens_details.cached_tokens marks the cached portion, which is billed at the cache rate, see Cache Billing.

Parsing Recommendations

• Always read choices[0].message.content; do not assume choices has more than 1 item;
• Reasoning models (-thinking tier) may include a reasoning_content field; when displaying, distinguish it from the main body, see Reasoning Model Output;
• When an error occurs, the response is in the {"error": {...}} structure; see Error Codes and Retries for handling.