Gpt5 Crewai issues

CrewAI Community Support
1

Hi I’m struggling to get GPT 5 to work in CrewAi tried both through OpenAI directly in the model through openrouter. GPT5 fails in both. the crew it to work with GPT-5-chat but not with the direct API, however the chat doesn’t allow function calling and agent delegation (which I believe is a tool call anyway).

Are there any default (required) parameters to set up the GPT-5 model model in CrewAI does anybody have some working examples of the parameters used to get GPT five working?

I’ve also tried to have a different model For function calling instantiated in the agents but this also seems to give an error, for some reason the agent seems to default to the overall LLM, an not revert to the function_calling_llm for tool calls. This seems to be especially true with openrouter

1 Like
2 
from crewai import LLM
import os

os.environ["OPENAI_API_KEY"] = "sk-YOUR-KEY"

gpt_5_llm = LLM(
    model="openai/gpt-5",
    drop_params=True,
    additional_drop_params=["stop", "temperature"]
)
3

Awesome. Thanks so much @maxmoura I’ll give it a try.

4

Thanks for this Max… For me GPT-5 is not great… I prefer 4.1

1 Like
5

Thanks Max, it worked, but I’m finding gpt-5 quite slow and agree that 4.1 seems better. I may try to see if I can play around with reasoning, verbosity and preamble to see if things improve. I might also need to update the agent and task prompts for 5 vs 4.1. I was also using hierarchical mode, I’ll try a sequential crew to see if there are any improvements.

Cheers, and thanks again. !

2 Likes
6

You know @danglies I am finding it really slow too.. I think I will wait for them to fix the problems before I put it near any production Crew.
Wondering: Is the slowness to do with OpenAI or as it is a new model the LLM manager in CrewAI not knowing how to work with it

7

This topic was automatically closed 24 hours after the last reply. New replies are no longer allowed.