mirror of
https://github.com/open-webui/open-webui.git
synced 2026-08-30 17:25:30 -05:00
issue: thinking not showing up for Openrouter models anymore after recent update #5412
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: github-starred/open-webui#5412
Reference in New Issue
Block a user
Originally created by @amanat361 on GitHub (May 31, 2025).
Check Existing Issues
Installation Method
Git Clone
Open WebUI Version
0.6.13
Ollama Version (if applicable)
No response
Operating System
Ubuntu 22.04
Browser (if applicable)
Chrome
Confirmation
README.md.Expected Behavior
View the "thinking" preview for reasoning models like Gemini 2.5 Pro. This worked normally for a while but upon the recent update no longer works. Local models on Ollama show reasoning, but external models do not. The expected behavior is the little spinner that says "thinking" where you can expand to see the reasoning tokens before the output completes.
Actual Behavior
Current behavior is that the model just has a loading state until reasoning completes and then continues with a standard response. For Deepseek models on Groq, it works, but for any Openrouter model (Gemini 2.5 pro or Deepseek R1 0528), it does not work. It also does not work for models from the Openai provider like O3 and O4-mini.
Steps to Reproduce
Clean Openwebui install, newest version. Ollama version irrelevant since this is an external provider issue. Go to "connections" tab in settings and add Openrouter and Openai providers respectively. Try models like Gemini 2.5 Pro
which has the tag
Openrouter.google/gemini-2.5-pro-previewSubmit a prompt in a chat that requires reasoning like "give me an interesting detailed proof for the square root of pi being irrational"
Notice that the thinking doesn't actually show. Sorry guys, I know the issue template wants me to put lots of info, but it's really as simple as that. On some models, reasoning doesn't show up. Not sure what else to explain!
Logs & Screenshots
Ok this is weird. When I went to take a screenshot, it literally worked the ONE time I tried to show it not working. Absolutely baffled.
However that was just Deepsek. I switched models to Gemini 2.5 Pro and yep, as expected, just a loading state:
In the browser console logs, the "delta" in the choice completion streaming is just an empty string over and over during the supposed reasoning stage:
Additional Information
Maybe it is a Gemini issue and the models changed or something but I could swear I remember the thinking working before, maybe I'm crazy. Let me know if you'd like more detailed info, I hope this is alright!
@tjbck commented on GitHub (Jun 2, 2025):
This might have more to do with model's stochastic behaviour than Open WebUI, we haven't changed our logic to handle reasoning content. Keep us updated!