AI Assistant Stuck "Thinking" for 35+ Mins: Charged Credits for Cancelled Executions?

Questions
1

Describe the problem/error/question

Sometimes when I use the built-in AI platform assistant to build or modify my workflows, it gets stuck in an infinite “Thinking” loop. In my latest run, it was stuck for over 35 minutes without any progress or output.
Because it was hanging for too long, I had to manually cancel/stop the execution. However, I discovered that my AI Credits were still deducted/charged for this run, despite getting absolutely zero output or change to my workflow.
Are we supposed to be charged for failed/hung AI assistant executions that we are forced to manually stop? Is there a way to prevent this freeze or get a credit refund for these lost runs?

What is the error message (if any)?

Please share your workflow

N/A (This is a platform-level issue with the built-in AI assistant interface, not a specific workflow node error).

Share the output returned by the last node

No output was returned because the AI assistant was manually stopped due to being permanently stuck.

Information on your n8n setup

    1. n8n version: Latest Cloud Version
    2. Database (default: SQLite): Default Cloud
    3. n8n EXECUTIONS_PROCESS setting: Default
    4. Running n8n via: n8n Cloud
    5. Operating system: Web Browser
2

This really isn’t something the community can verify or fix — only n8n support can see the actual token/execution log for that run and confirm whether the charge matches real usage. I’d recommend opening a support ticket (via the in-app support widget or your Cloud dashboard) with the execution ID and timestamp so they can look into it directly.

Worth noting — there’s another report of this exact “stuck Thinking + credits still charged” issue from a few days ago, so it might be a known bug rather than expected behavior. That said, whether you get credits back is really their call to make, not something I can promise from here.

3

Seconding the support ticket route — that’s the only path to get the credits reviewed.

One thing worth including in your ticket: the exact execution ID (visible in the Executions list even for cancelled runs) plus a screenshot of the credits usage log from your Cloud dashboard showing the charge timestamp. That gives support enough to cross-reference on their end without back-and-forth.

This does look like a known edge case where the AI assistant node holds the execution open waiting for a model response that never completes, and the credit deduction fires on execution start rather than on successful output. Flagging it with specifics helps the team track frequency.