Skip to main content
JSON Mode is a Pro feature. Upgrade to access.
Force the model to output valid, parseable JSON. Perfect for building structured data pipelines, APIs, and automated workflows.

​
Enabling JSON Mode

Set response_format to {"type": "json_object"} in your request:
Python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.llm.kiwi/v1",
    api_key="YOUR_API_KEY"
)

response = client.chat.completions.create(
    model="gpt-oss-20b",
    response_format={"type": "json_object"},
    messages=[
        {"role": "system", "content": "Extract data as JSON."},
        {"role": "user", "content": "Parse: John Doe, 30 years old, engineer"}
    ]
)

print(response.choices[0].message.content)
# {"name": "John Doe", "age": 30, "occupation": "engineer"}

​
Request Format

ParameterTypeDescription
response_formatobjectSet to {"type": "json_object"}
messagesarrayMust include “JSON” in system or user message
You must mention “JSON” in your prompt (system or user message). Otherwise, the request may fail or produce unexpected results.

​
Example Use Cases

​
Data Extraction

System Prompt
You are a data extraction assistant. Always respond with valid JSON matching this schema:
{
  "entities": [{"name": "string", "type": "string"}],
  "sentiment": "positive|negative|neutral",
  "summary": "string"
}

​
API Response Generation

Python
response = client.chat.completions.create(
    model="gpt-oss-20b",
    response_format={"type": "json_object"},
    messages=[
        {
            "role": "system",
            "content": """Generate API response JSON with structure:
            {"status": "success|error", "data": {...}, "message": "string"}"""
        },
        {"role": "user", "content": "Create a response for user signup success"}
    ]
)

​
Structured Analysis

Python
response = client.chat.completions.create(
    model="gpt-oss-20b",
    response_format={"type": "json_object"},
    messages=[
        {
            "role": "system",
            "content": "Analyze text and return JSON with keys: topics, keywords, reading_level"
        },
        {"role": "user", "content": "Analyze this article..."}
    ]
)

​
Model Compatibility

ModelJSON Mode Support
pro✅ Full support
fast✅ Full support
default⚠️ Best effort

​
Error Handling

If JSON parsing fails, wrap your response handling:
Python
import json

try:
    data = json.loads(response.choices[0].message.content)
except json.JSONDecodeError as e:
    # Handle invalid JSON
    print(f"Failed to parse: {e}")

​
Best Practices

  1. Define Schema in Prompt: Clearly specify the expected JSON structure
  2. Use System Messages: Put JSON requirements in the system role
  3. Validate Output: Always validate parsed JSON against your schema
  4. Handle Edge Cases: Account for optional fields and type variations

Need More Structure?

For even stricter schemas, use Function Calling with typed parameters.