mirror of
https://github.com/open-webui/open-webui.git
synced 2026-08-30 17:25:30 -05:00
feat: Add option in admin panel to set new API parameter for reasoning in GPT-5 #5990
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: github-starred/open-webui#5990
Reference in New Issue
Block a user
Originally created by @karolkt1 on GitHub (Aug 8, 2025).
Check Existing Issues
Problem Description
According to OpenAI docs, GPT-5 supports a reasoning parameter to control reasoning effort.
Currently, there is no way in OpenWebUI to pass this parameter to the API, which means we can't set it to "minimal" to improve response speed, by default, GPT-5 can be extremely slow without this control.
Desired Solution you'd like
New parametr in admin menu where the red arrow is
Alternatives Considered
No response
Additional Context
No response
@rgaricano commented on GitHub (Aug 8, 2025):
Right now, you can use custom parameter (at the end ot advanced params)
@karolkt1 commented on GitHub (Aug 8, 2025):
Could you help me how excatly because I tried some variations of reasoning/effort and always got errors. Other parameters visible on screenshot work just fine
@17jmumford commented on GitHub (Aug 8, 2025):
Reasoning effort is not available on the /chat/completions endpoints. This is true for both OpenAI and LiteLLM.
https://docs.litellm.ai/docs/completion/input
It's only available on the new Responses API. You would have to use Open WebUI's fancy pipeline/valve stuff to get it to work
@jsweetzer-ea commented on GitHub (Aug 8, 2025):
Reasoning effort is available in both. See here:
https://platform.openai.com/docs/guides/latest-model#migrating-from-chat-completions-to-responses-api
curl --request POST
--url https://api.openai.com/v1/chat/completions
--header "Authorization: Bearer $OPENAI_API_KEY"
--header 'Content-type: application/json'
--data '{
"model": "gpt-5",
"messages": [
{
"role": "user",
"content": "How much gold would it take to coat the Statue of Liberty in a 1mm layer?"
}
],
"reasoning_effort": "minimal"
}'
@karolkt1 commented on GitHub (Aug 9, 2025):
I do confirm that it works in /chat/completions. I tried 2 curls with minimal and high. The difference was almost 5 times longer response time with high reasoning.
So then I hope developers can add "reasoning" in right place of a call.
Below successful call with minimal reasoning and fast asnwer
@decent-engineer-decent-datascientist commented on GitHub (Aug 11, 2025):
Any updates on this? I too was not able to leverage reasoning_effort in the params.
@gaby commented on GitHub (Aug 11, 2025):
Reasoning effort is already there below
top_pin the model parameters.It defaults to "medium" as a string.
@karolkt1 commented on GitHub (Aug 11, 2025):
It doesn't work and it was the first thing I've tested. Try enabling it and then send request to GPT-5. I receive following response
@karolkt1 commented on GitHub (Aug 12, 2025):
If you are also using LiteLLM it was fixed in https://github.com/BerriAI/litellm/releases/tag/litellm_v1.75.5-dev_memory_fix