Mermaid Output Often Misses Quotes

ChatGPT Bugs
1

I’ve noticed on a few occasions that the mermaid output from o4-mini-high will contain invalid content, mostly related to unescaped labels containing parens which ought to be wrapped in double quotes.

failure:

flowchart TD
  A[Start] --> B{Need multimodal input<br/>(images, voice)?}
  
  B -->|Yes| C{Resource constraints?}
  C -->|Minimal| D[GPT-4o mini]
  C -->|Ample| E[GPT-4o]
  
  B -->|No| F{Require top-tier reasoning<br/>or very complex tasks?}
  F -->|Yes| G[GPT-4.5]
  F -->|No| H{Balance cost & capability?}
  H -->|High complexity,<br/>greater context| I[o4-mini-high]
  H -->|General purpose| J[o4-mini]
  H -->|Quick drafts,<br/>very low cost| K[o3]

success example:

flowchart TD
  A[Start] --> B{"Need multimodal input<br/>(images, voice)?"}
  
  B -->|Yes| C{Resource constraints?}
  C -->|Minimal| D[GPT-4o mini]
  C -->|Ample| E[GPT-4o]
  
  B -->|No| F{Require top-tier reasoning<br/>or very complex tasks?}
  F -->|Yes| G[GPT-4.5]
  F -->|No| H{Balance cost & capability?}
  H -->|High complexity,<br/>greater context| I[o4-mini-high]
  H -->|General purpose| J[o4-mini]
  H -->|Quick drafts,<br/>very low cost| K[o3]

I wonder if a pass could be done by just double checking mermaid output renders without errors before returning it to the user. If that kind of control is unavailable, the model requires more tuning around mermaid syntax.

/share/68117346-444c-8001-93be-0302ec4f898b