Addition of OpenAI new "developer" message type

[burakcbdn]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Yesterday openai released new type of message called "developer" message which replaces system message for newest models.

Langchain does not have a current solution for dynamically adding new message types, we should add new message type "developer" and make it possible to add dynamic types for future.

Motivation

Feature request for new technologies

Proposal (If applicable)

No response

[naofumi1014-ds]

I agree your request. We need developer message on langchain.

if you want to use markdown output ,we can use system role and content is Formatting re-enabled now.
But I recognize I should use developer role.

prompt = ChatPromptTemplate.from_messages([("system","Formatting re-enabled"), ("human","XXXXXX") ])

[phfifofum]

Yep I need this too. Is there any short term fix - anyone got a decent workaround for this?

[m4a1carbin4]

In My case of hotfix.
we can use system message type as temporal Developer messages.

from langchain_core.messages import SystemMessage, HumanMessage
from langchain_openai import ChatOpenAI

test_developer_message = SystemMessage(content="you are a developer")

test_developer_message.additional_kwargs = {"__openai_role__": "developer"}

test_message = HumanMessage(content="say somthing")

test_model = ChatOpenAI(model="o3-mini-2025-01-31", temperature=1, timeout=180)

print(test_model.invoke([test_developer_message, test_message]))

content="Hello there! I'm here and ready to help. What's on your mind today?" additional_kwargs={'refusal': None} response_metadata={'token_usage': {'completion_tokens': 410, 'prompt_tokens': 17, 'total_tokens': 427, 'completion_tokens_details': {'accepted_prediction_tokens': 0, 'audio_tokens': 0, 'reasoning_tokens': 384, 'rejected_prediction_tokens': 0}, 'prompt_tokens_details': {'audio_tokens': 0, 'cached_tokens': 0}}, 'model_name': 'o3-mini-2025-01-31', 'system_fingerprint': 'fp_8bcaa0ca21', 'finish_reason': 'stop', 'logprobs': None} id='****' usage_metadata={'input_tokens': 17, 'output_tokens': 410, 'total_tokens': 427, 'input_token_details': {'audio': 0, 'cache_read': 0}, 'output_token_details': {'audio': 0, 'reasoning': 384}}

somthing like this.

[m4a1carbin4]

for someone find hotfix's

we can use system message type as temporal Developer messages.

from langchain_core.messages import SystemMessage, HumanMessage
from langchain_openai import ChatOpenAI

test_developer_message = SystemMessage(content="you are a developer")

test_developer_message.additional_kwargs = {"__openai_role__": "developer"}

test_message = HumanMessage(content="say somthing")

test_model = ChatOpenAI(model="o3-mini-2025-01-31", temperature=1, timeout=180)

print(test_model.invoke([test_developer_message, test_message]))

content="Hello there! I'm here and ready to help. What's on your mind today?" additional_kwargs={'refusal': None} response_metadata={'token_usage': {'completion_tokens': 410, 'prompt_tokens': 17, 'total_tokens': 427, 'completion_tokens_details': {'accepted_prediction_tokens': 0, 'audio_tokens': 0, 'reasoning_tokens': 384, 'rejected_prediction_tokens': 0}, 'prompt_tokens_details': {'audio_tokens': 0, 'cached_tokens': 0}}, 'model_name': 'o3-mini-2025-01-31', 'system_fingerprint': 'fp_8bcaa0ca21', 'finish_reason': 'stop', 'logprobs': None} id='****' usage_metadata={'input_tokens': 17, 'output_tokens': 410, 'total_tokens': 427, 'input_token_details': {'audio': 0, 'cache_read': 0}, 'output_token_details': {'audio': 0, 'reasoning': 384}}