# OpenAI GPT-5 Issues

**URL:** <https://community.crewai.com/t/openai-gpt-5-issues/7005>\
**Category:** General\
**Created:** [September 4, 2025, 11:57am UTC](https://community.crewai.com/t/openai-gpt-5-issues/7005 "2025-09-04T11:57:03Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Tony\_Wood](https://sea1.discourse-cdn.com/flex025/user_avatar/community.crewai.com/tony_wood/32/1876_2.png) [@Tony\_Wood](https://community.crewai.com/u/Tony_Wood)\
**Post date:** [September 4, 2025, 11:57am UTC](https://community.crewai.com/t/openai-gpt-5-issues/7005/1 "2025-09-04T11:57:03Z")

</div>

I am still having issues with GPT 5.

I get the following messages

`Error parsing LLM output, agent will retry: I did it wrong. Tried to both perform Action and give a Final Answer at the same time, I must do one or the other`  
`Error parsing LLM output, agent will retry: I did it wrong. Tried to both perform Action and give a Final Answer at the same time, I must do one or the other`  
`Error parsing LLM output, agent will retry: I did it wrong. Tried to both perform Action and give a Final Answer at the same time, I must do one or the other`  
`Error parsing LLM output, agent will retry: I did it wrong. Tried to both perform Action and give a Final Answer at the same time, I must do one or the other`  
`Error parsing LLM output, agent will retry: I did it wrong. Tried to both perform Action and give a Final Answer at the same time, I must do one or the other`

I am using

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

When i use 4.1 or Gemini it works. For CrewAI I have stopped using GPT 5.. Any wisdom from the community for resolving this.. or is it just to wait for it to get a solid LLM implementation?

---

<div class="post-metadata">

**Author:** ![maxmoura](https://sea1.discourse-cdn.com/flex025/user_avatar/community.crewai.com/maxmoura/32/4206_2.png) [@maxmoura](https://community.crewai.com/u/maxmoura)\
**Post date:** [September 4, 2025, 4:20pm UTC](https://community.crewai.com/t/openai-gpt-5-issues/7005/2 "2025-09-04T16:20:33Z")

</div>

> [@Tony\_Wood](#):
>
> ```plaintext
> I did it wrong. Tried to both perform Action and give a Final Answer
> at the same time, I must do one or the other
> 
> ```

Well, judging by the output, you’re likely running into an issue I’ve already discussed [in this issue](https://github.com/crewAIInc/crewAI/issues/3154#issuecomment-3074633099). It comes down to the unnecessarily confusing way CrewAI presents the action space for the LLM to decide its next step. This can cause confusion, leading the LLM to provide a final answer when it shouldn’t.

If you have a moment, I’d recommend reading the entire thread to get the full picture. Further down in that same thread, I even put forward a simple proof-of-concept demonstrating how a single, clearer instruction can enable even an LLM with far fewer parameters to comprehend, decide, and communicate with clarity.

---

<div class="post-metadata">

**Author:** ![Tony\_Wood](https://sea1.discourse-cdn.com/flex025/user_avatar/community.crewai.com/tony_wood/32/1876_2.png) [@Tony\_Wood](https://community.crewai.com/u/Tony_Wood)\
**Post date:** [September 4, 2025, 4:45pm UTC](https://community.crewai.com/t/openai-gpt-5-issues/7005/3 "2025-09-04T16:45:14Z")

</div>

Hey Thanks Max. I’ll wait for the fix and switch to 4.1 which works well till then