Microsoft Foundry Series (Part 3) — Model Catalog: GPT-4.1 vs GPT-5 vs Open-Weight Models, Benchmarks, Pricing and Guidance

 1                         ┌───────────────────────────────┐
 2                         │  What's your primary need?     │
 3                         └──────┬────────────┬───────┬───┘
 4                                │            │       │
 5                          Reasoning    Coding/Long   Multi-modal
 6                          & complex    context       (audio/image)
 7                          tasks                     
 8                                │            │       │
 9                                ▼            ▼       ▼
10                            GPT-5       GPT-4.1    GPT-5 or
11                                                   GPT-4o
12                                │            │
13                     ┌──────────┴──┐  ┌──────┴──────┐
14                     │             │  │             │
15                High budget   Budget  High volume  Budget
16                     │        aware   │            aware
17                     ▼             │  ▼                │
18                  GPT-5            ▼  GPT-4.1          ▼
19                              GPT-5   (full)      GPT-4.1
20                              mini                 mini/nano

 1@description('Models to deploy')
 2param models array = [
 3  {
 4    name: 'gpt-4.1-mini'
 5    version: '2025-04-14'
 6    sku: 'GlobalStandard'
 7    capacity: 30
 8  }
 9  {
10    name: 'gpt-4.1'
11    version: '2025-04-14'
12    sku: 'GlobalStandard'
13    capacity: 10
14  }
15  {
16    name: 'gpt-5-mini'
17    version: '2025-06-02'
18    sku: 'GlobalStandard'
19    capacity: 10
20  }
21]

 1@batchSize(1) // deploy sequentially to avoid race conditions
 2resource deployments 'Microsoft.CognitiveServices/accounts/deployments@2025-04-01-preview' = [
 3  for model in models: {
 4    parent: foundry
 5    name: model.name
 6    sku: {
 7      name: model.sku
 8      capacity: model.capacity
 9    }
10    properties: {
11      model: {
12        name: model.name
13        format: 'OpenAI'
14        version: model.version
15      }
16    }
17  }
18]

1output deploymentNames array = [for (model, i) in models: deployments[i].name]

 1using 'main.bicep'
 2
 3param baseName = 'foundry-demo'
 4param location = 'swedencentral'
 5param models = [
 6  {
 7    name: 'gpt-4.1-mini'
 8    version: '2025-04-14'
 9    sku: 'GlobalStandard'
10    capacity: 30
11  }
12  {
13    name: 'gpt-5-mini'
14    version: '2025-06-02'
15    sku: 'GlobalStandard'
16    capacity: 10
17  }
18]

 1import os
 2import json
 3from dotenv import load_dotenv
 4from azure.identity import DefaultAzureCredential
 5from azure.ai.projects import AIProjectClient
 6
 7load_dotenv()
 8
 9project = AIProjectClient(
10    endpoint=os.environ["PROJECT_ENDPOINT"],
11    credential=DefaultAzureCredential(),
12)
13openai = project.get_openai_client()
14
15# Model routing map
16MODELS = {
17    "simple": "gpt-4.1-nano",     # classification, extraction
18    "standard": "gpt-4.1-mini",   # product descriptions, summaries
19    "complex": "gpt-5-mini",      # multi-step reasoning, planning
20}
21
22
23def classify_complexity(task: str) -> str:
24    """Use the cheapest model to classify task complexity."""
25    response = openai.responses.create(
26        model=MODELS["simple"],
27        instructions=(
28            "Classify the following task as 'simple', 'standard', or 'complex'. "
29            "Return only the classification word."
30        ),
31        input=task,
32    )
33    classification = response.output_text.strip().lower()
34    return classification if classification in MODELS else "standard"
35
36
37def execute_task(task: str, system_prompt: str) -> str:
38    """Route to the appropriate model based on complexity."""
39    complexity = classify_complexity(task)
40    model = MODELS[complexity]
41
42    print(f"Task complexity: {complexity} → using {model}")
43
44    response = openai.responses.create(
45        model=model,
46        instructions=system_prompt,
47        input=task,
48    )
49    return response.output_text
50
51
52if __name__ == "__main__":
53    # Simple task → routed to nano
54    print(execute_task(
55        task="Classify this product as Electronics, Clothing, or Home: 'Wireless Bluetooth Headphones'",
56        system_prompt="You are a product classifier. Return only the category name.",
57    ))
58
59    print("---")
60
61    # Standard task → routed to mini
62    print(execute_task(
63        task="Write a product description for 'TrailBlazer Pro Hiking Boots' made of leather with Vibram soles",
64        system_prompt="You are an e-commerce copywriter. Write 2-3 compelling sentences.",
65    ))
66
67    print("---")
68
69    # Complex task → routed to GPT-5 mini
70    print(execute_task(
71        task="Analyse our product catalog and suggest a pricing strategy that maximises revenue while maintaining competitiveness. Consider seasonal trends, competitor pricing, and customer segments.",
72        system_prompt="You are a retail pricing strategist. Provide a structured analysis with recommendations.",
73    ))

 1Task complexity: simple → using gpt-4.1-nano
 2Electronics
 3---
 4Task complexity: standard → using gpt-4.1-mini
 5Conquer any terrain with the TrailBlazer Pro Hiking Boots, crafted from premium 
 6leather for durability that lasts. Equipped with Vibram soles for unmatched grip 
 7on rocky trails, these boots are your ultimate companion for every adventure.
 8---
 9Task complexity: complex → using gpt-5-mini
10## Pricing Strategy Analysis
11...

1az group delete --name rg-foundry-demo --yes --no-wait