Structured Outputs

Structured Output Schema

You can set the response_format parameter to your defined schema to ensure the model produces a JSON object that matches your specified structure.

1import openai
2import json
3
4# Define the OpenAI client
5client = openai.OpenAI(
6    base_url="https://api.scx.ai/v1", 
7    api_key="your-scx-api-key"
8)
9
10MODEL = "DeepSeek-V3-0324"
11
12# Define the schema
13response_format = {
14    "type": "json_schema",
15    "json_schema": {
16        "name": "data_extraction",
17        "schema": {
18            "type": "object",
19            "properties": {
20                "section": {
21                    "type": "string"
22                },
23                "products": {
24                    "type": "array",
25                    "items": {
26                        "type": "string"
27                    }
28                }
29            },
30            "required": ["section", "products"],
31            "additionalProperties": False
32        },
33        "strict": False 
34    }
35}
36
37# Call the API
38completion = client.chat.completions.create(
39    model=MODEL,
40    messages=[
41        {
42            "role": "system",
43            "content": "You are an expert at structured data extraction. You will be given unstructured text and should convert it into the given structure."
44        },
45        {
46            "role": "user",
47            "content": "the section 24 has appliances, and videogames"
48        }
49    ],
50    response_format=response_format
51)
52
53# Print the parsed result
54print(completion)
55
56

1import openai
2import json
3
4# Define the OpenAI client
5client = openai.OpenAI(
6    base_url="https://api.scx.ai/v1", 
7    api_key="your-scx-api-key"
8)
9
10MODEL = "DeepSeek-V3-0324"
11
12# Define the schema
13response_format = {
14    "type": "json_schema",
15    "json_schema": {
16        "name": "data_extraction",
17        "schema": {
18            "type": "object",
19            "properties": {
20                "section": {
21                    "type": "string"
22                },
23                "products": {
24                    "type": "array",
25                    "items": {
26                        "type": "string"
27                    }
28                }
29            },
30            "required": ["section", "products"],
31            "additionalProperties": False
32        },
33        "strict": False 
34    }
35}
36
37# Call the API
38completion = client.chat.completions.create(
39    model=MODEL,
40    messages=[
41        {
42            "role": "system",
43            "content": "You are an expert at structured data extraction. You will be given unstructured text and should convert it into the given structure."
44        },
45        {
46            "role": "user",
47            "content": "the section 24 has appliances, and videogames"
48        }
49    ],
50    response_format=response_format
51)
52
53# Print the parsed result
54print(completion)
55
56

Note

Ensure to set the "strict" parameter to false, as true isn’t supported yet. When it is available, it will ensure the model strictly follows your function schema instead of making a best-effort attempt.

JSON mode

You can set the response_format parameter to json_object in your request to ensure that the model outputs a valid JSON. In case the mode is not able to generate a valid JSON, an error will be returned.

Note

In case the model fails to generate a valid JSON, you will get an error message Model did not output valid JSON.

1import openai
2
3# Define the OpenAI client
4client = openai.OpenAI(
5    base_url="https://api.scx.ai/v1", 
6    api_key="your-scx-api-key"
7)
8
9MODEL = 'Meta-Llama-3.3-70B-Instruct'
10
11
12def run_conversation(user_prompt):
13    # Initial conversation with user input
14    messages = [
15        {
16            "role": "system",
17            "content": "Always provide the response in this JSON format: {\"country\": \"name\", \"capital\": \"xx\"}"
18        },
19
20        {
21            "role": "user",
22            "content": user_prompt,
23        }
24    ]
25
26    # First API call to get model's response
27    response = client.chat.completions.create(
28        model=MODEL,
29        messages=messages,
30        max_tokens=500,
31        response_format = { "type": "json_object"},
32        # stream = True
33    )
34    
35    response_message = response.choices[0].message
36    print(response_message)
37
38
39run_conversation('what is the capital of Austria')
40

1import openai
2
3# Define the OpenAI client
4client = openai.OpenAI(
5    base_url="https://api.scx.ai/v1", 
6    api_key="your-scx-api-key"
7)
8
9MODEL = 'Meta-Llama-3.3-70B-Instruct'
10
11
12def run_conversation(user_prompt):
13    # Initial conversation with user input
14    messages = [
15        {
16            "role": "system",
17            "content": "Always provide the response in this JSON format: {\"country\": \"name\", \"capital\": \"xx\"}"
18        },
19
20        {
21            "role": "user",
22            "content": user_prompt,
23        }
24    ]
25
26    # First API call to get model's response
27    response = client.chat.completions.create(
28        model=MODEL,
29        messages=messages,
30        max_tokens=500,
31        response_format = { "type": "json_object"},
32        # stream = True
33    )
34    
35    response_message = response.choices[0].message
36    print(response_message)
37
38
39run_conversation('what is the capital of Austria')
40

1ChatCompletionMessage(content='{"country": "Austria", "capital": "Vienna"}', role='assistant', function_call=None, tool_calls=None)
2

1ChatCompletionMessage(content='{"country": "Austria", "capital": "Vienna"}', role='assistant', function_call=None, tool_calls=None)
2

Other methods of structured outputs

Beyond JSON mode, structured outputs can be generated using the Instructor library. Learn more on the Instructor integration page.