When the model (qwen/qwen3.5-35b-a3b) uses reasoning tokens and tool calls (e.g., fetch_url), the response is rendered as raw text including all XML-like tags instead of being parsed and displayed in a structured UI.
Current behavior
The chat displays raw markup in the message body:
<details type="reasoning" done="true" duration="0"> <summary>Thought for 0 seconds</summary> ... shown as plain text<details type="tool_calls" done="true" id="821309517" name="fetch_url" arguments="..."> rendered literally- Raw
result=""..." strings with escaped quotes and \n characters visible to the user
- The entire tool call lifecycle (arguments, raw HTTP response body) is dumped into the chat
Expected behavior
Reasoning blocks should collapse into a "Thought for X seconds" toggle (like Open WebUI does — see screenshots). Tool calls should render as a collapsible "Explored 2 fetch_url" summary with expandable details, not as raw serialized text.
Screenshots
| Bug (current) |
Expected (Open WebUI) |
 |
 |
 |
 |
Environment
- Model:
qwen/qwen3.5-35b-a3b (via Ollama)
- OS: iOS
Additional context
Open WebUI correctly parses these same <details> blocks and tool_calls metadata into collapsible UI elements. The issue appears to be that the client is not parsing the structured output format that Qwen 3.5 emits for reasoning and tool use.
When the model (qwen/qwen3.5-35b-a3b) uses reasoning tokens and tool calls (e.g.,
fetch_url), the response is rendered as raw text including all XML-like tags instead of being parsed and displayed in a structured UI.Current behavior
The chat displays raw markup in the message body:
<details type="reasoning" done="true" duration="0"> <summary>Thought for 0 seconds</summary> ...shown as plain text<details type="tool_calls" done="true" id="821309517" name="fetch_url" arguments="...">rendered literallyresult=""..."strings with escaped quotes and\ncharacters visible to the userExpected behavior
Reasoning blocks should collapse into a "Thought for X seconds" toggle (like Open WebUI does — see screenshots). Tool calls should render as a collapsible "Explored 2 fetch_url" summary with expandable details, not as raw serialized text.
Screenshots
Environment
qwen/qwen3.5-35b-a3b(via Ollama)Additional context
Open WebUI correctly parses these same
<details>blocks andtool_callsmetadata into collapsible UI elements. The issue appears to be that the client is not parsing the structured output format that Qwen 3.5 emits for reasoning and tool use.