[Qwen3 32B Model - Groq API] 'Thinking' Section Should Be Collapsible in Agent Mode and When Copying Output

[lutfi-zain]

Description

Summary
When using the Qwen model via Groq API in OpenWhisper:

• In Agent Mode, the "Thinking" section is included in the output but should be collapsible (optional to view).
  Currently, the "Thinking" section is always visible and copied, which is inconsistent with user expectations.

---

Steps to Reproduce

1. Agent Mode:
   • Use Qwen model via Groq API.
   • Observe output format: Thinking: [process] Output: [final answer].
   • Attempt to copy the output.
   • Issue: "Thinking" section is visible and copied.

---

Expected Behavior

• Agent Mode:
  • "Thinking" section is collapsible (e.g., a toggle button to hide/show it).
  • When copying, only the final answer is selected.

---

Actual Behavior

• "Thinking" section is always visible and copied in both modes.

---

Screenshots

---

Additional Context

• This behavior is specific to the Quen model.
• Other models (e.g., Llama, Mistral) do not exhibit this issue.
• Check if Grog API supports mode-specific output formatting (e.g., agent_mode=true or non_agent_mode=true).