Canvas Truncating Code Output at 48 Lines – Pro User, Still No Support Response

ChatGPT
,
1

Hi

I’m a ChatGPT Pro subscriber and I’ve been dealing with a serious issue in Canvas for over a week, with no resolution from support after 12+ hours of silence .

Here’s what’s happening:

When I paste in a code block (around 500 lines ) into Canvas and ask ChatGPT to refactor, debug, or rewrite it, the output gets cut off at exactly 48 lines —no warning, no error. It just silently stops.

To try and work around it, ChatGPT sometimes:

  • Offers a download link , which returns a file with literally line 1, line 2, line 3, repeated 500 times—not actual code .
  • Tries to output the response through the normal message stream instead of Canvas, but it still breaks formatting or truncates .

I’ve tested this across:

  • Two different Windows systems
  • Multiple browsers (Chrome, Edge)
  • Incognito mode with no extensions
  • Fully cleared cookies, cache, and storage
  • Switched networks (Wi-Fi, mobile)
  • Logged out of all devices
  • Deleted all chat history to reset session state

Nothing fixes it.

This appears to be a backend limitation or bug specifically affecting Canvas’ ability to return full-length code completions.

I rely on Canvas heavily in my workflow and it’s completely unusable in this state. I’ve already submitted full diagnostics and a HAR log to OpenAI support. But after a week of issues and now nearly half a day of radio silence , I’m reaching out here.

Has anyone else experienced this hard cutoff at 48 lines? Is there a known fix or flag restriction tied to Pro accounts that support hasn’t acknowledged?

Thanks in advance,

2

I’ve now run multiple tests outside of Canvas to isolate the issue.

  • I uploaded a 486-line .py file into ChatGPT
  • Asked the assistant to patch it with 10 new lines of code
  • It returned a clean file via download with 499 lines , and I received exactly 498 , which is expected (1 line difference from editor behavior)

That proves:
:white_check_mark: File uploads and downloads work perfectly
:white_check_mark: Manual assistant-generated code patches work
:cross_mark: Canvas still silently truncates or fails to display full code
:cross_mark: In one test, Canvas showed 101 lines, but backend confirmed only 54 were actually injected

This confirms the bug is isolated to Canvas —not file size, system, or formatting.

Still waiting on a response from OpenAI Support after over a week of impact and multiple escalation attempts.