Newline/Whitespace Corruption

Hi Venice team. I’m seeing consistent newline/whitespace corruption in the chat UI that appears independent of the model (reproduces with GLM 4.7, DeepSeek v3.2, and GPT-5.2).

Issue: When prompting a model to return text with exact line breaks (including blank lines, indentation, bullets) the output is often collapsed into a single line or otherwise reformatted. Even fenced code blocks sometimes lose internal newlines. Conversely, in some cases the UI also seems to inject extra blank lines (e.g., double-spacing every line).

Repro (simple):

1. Ask model: “Reply with exactly these 5 lines, preserving the blank line:” then provide:

   • LINE 1

   • LINE 2

   • (blank line)

   • LINE 4

   • END

2. Result in UI frequently collapses/normalizes line breaks.

Observed: Copy/pasting the displayed output into a plain text editor shows the line breaks are not preserved as expected, suggesting UI rendering or pre/post-processing is altering whitespace.