[BUG] CrewAI >= 1.4.0 - Structured outputs no longer working with azure models

[LouisLecouturier]

Description

After updating to crewai 1.4.1, I can no longer use structured outputs with my Azure models

I always get this error:

ERROR:root:Azure API call failed: Unsupported `response_format` {'type': 'json_schema', 'json_schema': {<MYJSONSCHEMA>}

Steps to Reproduce

1. Use crewai[azure-ai-inference] 1.4.1
2. Use an Azure model
3. Create an agent or call a Azure LLM with a pydantic model as response format

Expected behavior

It should just return the structured response as written in the docs

Screenshots/Code snippets

from crewai import LLM, Agent
from dotenv import load_dotenv
from pydantic import BaseModel

class Greeting(BaseModel):
    name: str
    message: str

GPT_4_1_NANO = LLM(
    temperature=0,
    model="azure/gpt-4.1-nano",
    api_key=os.getenv("AZURE_API_KEY"),
    endpoint=os.getenv("MYDEPLOYMENTURL"),
    max_retries=5,
)

greeting_agent = Agent(
    role="Greeter",
    goal="Greet the user with a friendly message.",
    backstory="An AI assistant that loves to greet people.",
    llm=GPT_4_1_NANO,
)

res = greeting_agent.kickoff("My name is Alice.", response_format=Greeting)
print("Agent Res:", res)

Same with this:

from crewai import LLM, Agent
from dotenv import load_dotenv
from pydantic import BaseModel

class Greeting(BaseModel):
    name: str
    message: str

GPT_4_1_NANO = LLM(
    temperature=0,
    model="azure/gpt-4.1-nano",
    api_key=os.getenv("AZURE_API_KEY"),
    endpoint=os.getenv("MYDEPLOYMENTURL"),
    max_retries=5,
)

res = GPT_4_1_NANO.call("My name is Alice.", response_model=Greeting)
print("Agent Res:", res)

Operating System

Ubuntu 20.04

Python Version

3.12

crewAI Version

1.4.1
"crewai[azure-ai-inference]>=1.4.1",

crewAI Tools Version

N.A

Virtual Environment

Venv

Evidence

Possible Solution

None

Additional context

N.A

[LouisLecouturier]

I rollbacked to crewAI 1.3.0 and the issue seems to be gone.

I had the issue with with versions:

• 1.4.1
• 1.4.0

[Vidit-Ostwal]

Hey @LouisLecouturier, is this not working with just gpt-4.1-nano or every azure hosted model?

[LouisLecouturier]

Hello @Vidit-Ostwal I am having this issue with all the azure models I was using
4.1, 4.1 mini, 4.1 nano, o4 mini

[Vidit-Ostwal]

Have patched it in the above PR.

[LouisLecouturier]

Have patched it in the above PR.

Thanks a lot ! When is the next release planned ?

[Vidit-Ostwal]

Have patched it in the above PR.

Thanks a lot ! When is the next release planned ?

Not sure, just a OSS contributor here, not from the official team :)

[Vidit-Ostwal]

Have patched it in the above PR.

Thanks a lot ! When is the next release planned ?

Mind testing this particular PR for me though
just do a

pip install "git+https://github.com/Vidit-Ostwal/crewAI.git@vo/fix/azure-ChatCompletionsClient#subdirectory=lib/crewai"

[LouisLecouturier]

@Vidit-Ostwal tested with this code:

class Greeting(BaseModel):
    name: str = Field(..., description="Name of the person to greet")
    message: str = Field(..., description="Greeting message")

llm = GPT_4_1_MINI
res = llm.call("Hey I am Alice, nice to meet you!", response_model=Greeting)

print(res)

Got:

---

WIth this code:

class Greeting(BaseModel):
    name: str = Field(..., description="Name of the person to greet")
    message: str = Field(..., description="Greeting message")


greeting_agent = Agent(
    role="Greeter",
    goal="Greet people in a friendly manner.",
    backstory="An AI assistant that loves to greet people warmly.",
    llm=GPT_4_1_MINI,
)

res = greeting_agent.kickoff("Hey i am Bob, nice to meet you!", response_format=Greeting)
print(res)

Got:

[Vidit-Ostwal]

Mind testing it again, you will have to
pip install "git+https://github.com/Vidit-Ostwal/crewAI.git@vo/fix/azure-ChatCompletionsClient#subdirectory=lib/crewai"

Didn't see the regex check on the name.

[LouisLecouturier]

A new error appeared @Vidit-Ostwal

Code:

def test_fix():
    llm = GPT_4_1_MINI

    class Greeting(BaseModel):
        name: str = Field(..., description="The name of the person")
        message: str = Field(..., description="The greeting message")

    res = llm.call("Hey, I am bob", response_model=Greeting)
    print(res)


    agent = Agent(
        role="Greeting agent",
        goal="Greets all newcomers",
        backstory="You are a friendly greeting agent that welcomes new users.",
        llm=llm,

    )

    res = agent.kickoff('Hey I am Alice', response_format=Greeting)
    print(res)

test_fix()

error: