LLM Model LandscapeFoundational

Build a Multimodal GPT App

The GPT-6 tiers, what reasoning effort replaced, when to use the Responses API over Chat Completions, how the bill is calculated, and what is different on Azure.

By Amit UpadhyayPublished 13 Oct 202611 min

0/1 steps

Build Steps

Progress

0/1

◆ Before you start

Prerequisites

Python 3.11+Required
python --version
openai >= 1.30Required
pip install openai
OPENAI_API_KEYRequired
export OPENAI_API_KEY=...

🧠

Step 1 of 1

Build with GPT-6: vision, JSON mode, and streaming

Use the GPT-6 family for vision, structured JSON output, streaming, and function calling.

Tasks

Send a text + image request to gpt-6-luna

Use json_object mode to extract structured data from text

Stream a long response and track TTFT vs total latency

Define 2 function tools and handle parallel tool calls

Use Structured Outputs with response_format=json_schema

🎯 What You'll Learn

▸Send images to a GPT-6 model for visual question answering
▸Use response_format json_object for structured output
▸Stream responses and measure TTFT

Code

Pythongpt4o_demo.py
1 import os
2 import json
3 import time
4 import openai
5 from pydantic import BaseModel
6  
7 client = openai.OpenAI(api_key=os.environ["OPENAI_API_KEY"])
8  
9 # ── 1. Structured Outputs with Pydantic ────────────────────────────
10 class TechStack(BaseModel):
11     languages: list[str]
12     frameworks: list[str]
13     databases: list[str]
14     cloud: list[str]
15  
16 response = client.beta.chat.completions.parse(
17     model="gpt-6-luna", reasoning_effort="none",
18     messages=[{
19         "role": "user",
20         "content": "What tech stack does Netflix likely use? List specific technologies.",
21     }],
22     response_format=TechStack,
23 )
24 stack = response.choices[0].message.parsed
25 print("=== Structured Output ===")
26 print(f"Languages: {stack.languages}")
27 print(f"Databases: {stack.databases}")
28  
29 # ── 2. Streaming with TTFT measurement ─────────────────────────────
30 print("\n=== Streaming ===")
31 start = time.perf_counter()
32 ttft = None
33 full_response = []
34  
35 stream = client.chat.completions.create(
36     model="gpt-6-luna", reasoning_effort="none",
37     messages=[{"role": "user", "content": "Write a short poem about machine learning."}],
38     max_completion_tokens=150,
39     stream=True,
40 )
41  
42 for chunk in stream:
43     delta = chunk.choices[0].delta.content
44     if delta:
45         if ttft is None:
46             ttft = (time.perf_counter() - start) * 1000
47         full_response.append(delta)
48         print(delta, end="", flush=True)
49  
50 total = (time.perf_counter() - start) * 1000
51 print(f"\nTTFT: {ttft:.0f}ms, Total: {total:.0f}ms")
52  
53 # ── 3. Parallel Function Calling ───────────────────────────────────
54 tools = [
55     {
56         "type": "function",
57         "function": {
58             "name": "get_stock_price",
59             "description": "Get current stock price",
60             "parameters": {
61                 "type": "object",
62                 "properties": {"ticker": {"type": "string"}},
63                 "required": ["ticker"],
64             },
65         },
66     },
67     {
68         "type": "function",
69         "function": {
70             "name": "get_news",
71             "description": "Get latest news for a company",
72             "parameters": {
73                 "type": "object",
74                 "properties": {"company": {"type": "string"}},
75                 "required": ["company"],
76             },
77         },
78     },
79 ]
80  
81 response = client.chat.completions.create(
82     model="gpt-6-luna", reasoning_effort="none",
83     messages=[{"role": "user", "content": "Get AAPL stock price and latest Apple news"}],
84     tools=tools,
85     tool_choice="auto",
86 )
87  
88 print("\n=== Parallel Tool Calls ===")
89 if response.choices[0].finish_reason == "tool_calls":
90     for tc in response.choices[0].message.tool_calls:
91         print(f"Tool: {tc.function.name}({tc.function.arguments})")
92

💡 Pro Tips

▸Structured Outputs (not just JSON mode) guarantees schema conformance — use it for extraction
▸gpt-6-luna handles most tasks at 1/20th the cost of gpt-6.1-sol, and with reasoning_effort="none" it still accepts tools and temperature on Chat Completions; Sol and Astra need the Responses API for tool calling

GPT-6 OpenAI reasoning-effort Responses-API model-selection

← PreviousClaude: Anthropic's Model Family in Practice

Continue reading

All LLM Model Landscape articles →

LLM Model Landscape11 min

Claude: Anthropic's Model Family in Practice

The four Claude tiers, adaptive thinking and effort, the tool-use rules that differ from other providers, prompt caching, and where you can run it.

LLM Model Landscape11 min

Llama: Open-Weight Models for Production

The Llama lineup and licence, serving with Ollama and vLLM, GPU memory sizing, and an honest look at when a hosted API is still the better choice.

LLM Model Landscape10 min

Evals for Choosing a Model

Why public benchmarks do not transfer to your workload, how to build a model-selection eval, and why cost per successful task beats price per token.

Discussion

1 import os 2 import json 3 import time 4 import openai 5 from pydantic import BaseModel 6 7 client = openai.OpenAI(api_key=os.environ["OPENAI_API_KEY"]) 8 9 # ── 1. Structured Outputs with Pydantic ──────────────────────────── 10 class TechStack(BaseModel): 11 languages: list[str] 12 frameworks: list[str] 13 databases: list[str] 14 cloud: list[str] 15 16 response = client.beta.chat.completions.parse( 17 model="gpt-6-luna", reasoning_effort="none", 18 messages=[{ 19 "role": "user", 20 "content": "What tech stack does Netflix likely use? List specific technologies.", 21 }], 22 response_format=TechStack, 23 ) 24 stack = response.choices[0].message.parsed 25 print("=== Structured Output ===") 26 print(f"Languages: {stack.languages}") 27 print(f"Databases: {stack.databases}") 28 29 # ── 2. Streaming with TTFT measurement ───────────────────────────── 30 print("\n=== Streaming ===") 31 start = time.perf_counter() 32 ttft = None 33 full_response = [] 34 35 stream = client.chat.completions.create( 36 model="gpt-6-luna", reasoning_effort="none", 37 messages=[{"role": "user", "content": "Write a short poem about machine learning."}], 38 max_completion_tokens=150, 39 stream=True, 40 ) 41 42 for chunk in stream: 43 delta = chunk.choices[0].delta.content 44 if delta: 45 if ttft is None: 46 ttft = (time.perf_counter() - start) * 1000 47 full_response.append(delta) 48 print(delta, end="", flush=True) 49 50 total = (time.perf_counter() - start) * 1000 51 print(f"\nTTFT: {ttft:.0f}ms, Total: {total:.0f}ms") 52 53 # ── 3. Parallel Function Calling ─────────────────────────────────── 54 tools = [ 55 { 56 "type": "function", 57 "function": { 58 "name": "get_stock_price", 59 "description": "Get current stock price", 60 "parameters": { 61 "type": "object", 62 "properties": {"ticker": {"type": "string"}}, 63 "required": ["ticker"], 64 }, 65 }, 66 }, 67 { 68 "type": "function", 69 "function": { 70 "name": "get_news", 71 "description": "Get latest news for a company", 72 "parameters": { 73 "type": "object", 74 "properties": {"company": {"type": "string"}}, 75 "required": ["company"], 76 }, 77 }, 78 }, 79 ] 80 81 response = client.chat.completions.create( 82 model="gpt-6-luna", reasoning_effort="none", 83 messages=[{"role": "user", "content": "Get AAPL stock price and latest Apple news"}], 84 tools=tools, 85 tool_choice="auto", 86 ) 87 88 print("\n=== Parallel Tool Calls ===") 89 if response.choices[0].finish_reason == "tool_calls": 90 for tc in response.choices[0].message.tool_calls: 91 print(f"Tool: {tc.function.name}({tc.function.arguments})") 92

1	import os
2	import json
3	import time
4	import openai
5	from pydantic import BaseModel
6
7	client = openai.OpenAI(api_key=os.environ["OPENAI_API_KEY"])
8
9	# ── 1. Structured Outputs with Pydantic ────────────────────────────
10	class TechStack(BaseModel):
11	languages: list[str]
12	frameworks: list[str]
13	databases: list[str]
14	cloud: list[str]
15
16	response = client.beta.chat.completions.parse(
17	model="gpt-6-luna", reasoning_effort="none",
18	messages=[{
19	"role": "user",
20	"content": "What tech stack does Netflix likely use? List specific technologies.",
21	}],
22	response_format=TechStack,
23	)
24	stack = response.choices[0].message.parsed
25	print("=== Structured Output ===")
26	print(f"Languages: {stack.languages}")
27	print(f"Databases: {stack.databases}")
28
29	# ── 2. Streaming with TTFT measurement ─────────────────────────────
30	print("\n=== Streaming ===")
31	start = time.perf_counter()
32	ttft = None
33	full_response = []
34
35	stream = client.chat.completions.create(
36	model="gpt-6-luna", reasoning_effort="none",
37	messages=[{"role": "user", "content": "Write a short poem about machine learning."}],
38	max_completion_tokens=150,
39	stream=True,
40	)
41
42	for chunk in stream:
43	delta = chunk.choices[0].delta.content
44	if delta:
45	if ttft is None:
46	ttft = (time.perf_counter() - start) * 1000
47	full_response.append(delta)
48	print(delta, end="", flush=True)
49
50	total = (time.perf_counter() - start) * 1000
51	print(f"\nTTFT: {ttft:.0f}ms, Total: {total:.0f}ms")
52
53	# ── 3. Parallel Function Calling ───────────────────────────────────
54	tools = [
55	{
56	"type": "function",
57	"function": {
58	"name": "get_stock_price",
59	"description": "Get current stock price",
60	"parameters": {
61	"type": "object",
62	"properties": {"ticker": {"type": "string"}},
63	"required": ["ticker"],
64	},
65	},
66	},
67	{
68	"type": "function",
69	"function": {
70	"name": "get_news",
71	"description": "Get latest news for a company",
72	"parameters": {
73	"type": "object",
74	"properties": {"company": {"type": "string"}},
75	"required": ["company"],
76	},
77	},
78	},
79	]
80
81	response = client.chat.completions.create(
82	model="gpt-6-luna", reasoning_effort="none",
83	messages=[{"role": "user", "content": "Get AAPL stock price and latest Apple news"}],
84	tools=tools,
85	tool_choice="auto",
86	)
87
88	print("\n=== Parallel Tool Calls ===")
89	if response.choices[0].finish_reason == "tool_calls":
90	for tc in response.choices[0].message.tool_calls:
91	print(f"Tool: {tc.function.name}({tc.function.arguments})")
92