[Bug]: Search memory failed with Qwen3.5-xxx thinking mode (default enabled)

Describe the bug

Thinking mode is enabled by default since Qwen3.5-xxx (qwen3.5-plus, qwen3.5-flash) and memmachine search memory could not handle section in Pydantic model_validate_json()

2026-03-25 03:05:22,890 [ERROR] memmachine_server.semantic_memory.semantic_ingestion - Failed to process message 5 for semantic type profile
Traceback (most recent call last):
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/semantic_memory/semantic_ingestion.py", line 163, in process_semantic_type
    commands = await llm_feature_update(
               ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/pydantic/_internal/_validate_call.py", line 34, in wrapper_function
    return await wrapper(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/semantic_memory/semantic_llm.py", line 88, in llm_feature_update
    parsed_output = await model.generate_parsed_response(
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/common/language_model/openai_responses_language_model.py", line 157, in generate_parsed_response
    response = await self._client.with_options(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/resources/responses/responses.py", line 2825, in parse
    return await self._post(
           ^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1884, in post
    return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1674, in request
    return await self._process_response(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1763, in _process_response
    return await api_response.parse()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_response.py", line 432, in parse
    parsed = self._options.post_parser(parsed)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/resources/responses/responses.py", line 2819, in parser
    return parse_response(
           ^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_responses.py", line 74, in parse_response
    "parsed": parse_text(item.text, text_format=text_format),
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_responses.py", line 139, in parse_text
    return cast(TextFormatT, model_parse_json(text_format, text))
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_compat.py", line 171, in model_parse_json
    return model.model_validate_json(data)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/pydantic/main.py", line 766, in model_validate_json
    return cls.__pydantic_validator__.validate_json(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
pydantic_core._pydantic_core.ValidationError: 1 validation error for _SemanticFeatureUpdateRes
  Invalid JSON: expected value at line 1 column 1 [type=json_invalid, input_value='<think>\nThe user is sha...atabase work"\n    }\n]', input_type=str]
    For further information visit https://errors.pydantic.dev/2.12/v/json_invalid
2026-03-25 03:05:45,040 [INFO] httpx - HTTP Request: POST https://dashscope.aliyuncs.com/compatible-mode/v1/responses "HTTP/1.1 200 OK"
2026-03-25 03:05:45,041 [ERROR] memmachine_server.semantic_memory.semantic_ingestion - Failed to process message 6 for semantic type profile
Traceback (most recent call last):
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/semantic_memory/semantic_ingestion.py", line 163, in process_semantic_type
    commands = await llm_feature_update(
               ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/pydantic/_internal/_validate_call.py", line 34, in wrapper_function
    return await wrapper(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/semantic_memory/semantic_llm.py", line 88, in llm_feature_update
    parsed_output = await model.generate_parsed_response(
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/common/language_model/openai_responses_language_model.py", line 157, in generate_parsed_response
    response = await self._client.with_options(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/resources/responses/responses.py", line 2825, in parse
    return await self._post(
           ^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1884, in post
    return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1674, in request
    return await self._process_response(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1763, in _process_response
    return await api_response.parse()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_response.py", line 432, in parse
    parsed = self._options.post_parser(parsed)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/resources/responses/responses.py", line 2819, in parser
    return parse_response(
           ^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_responses.py", line 74, in parse_response
    "parsed": parse_text(item.text, text_format=text_format),
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_responses.py", line 139, in parse_text
    return cast(TextFormatT, model_parse_json(text_format, text))
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_compat.py", line 171, in model_parse_json
    return model.model_validate_json(data)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/pydantic/main.py", line 766, in model_validate_json
    return cls.__pydantic_validator__.validate_json(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
pydantic_core._pydantic_core.ValidationError: 1 validation error for _SemanticFeatureUpdateRes
  Invalid JSON: expected value at line 1 column 1 [type=json_invalid, input_value='<think>\nLooking at this...Third Board)"\n    }\n]', input_type=str]
    For further information visit https://errors.pydantic.dev/2.12/v/json_invalid

Steps to reproduce

From Qwen official document, Thinking mode is enabled by default since qwen3.5

https://www.alibabacloud.com/help/en/model-studio/deep-thinking#19a4c1988fp6d

Expected behavior

Search memory may report such error and add memory works well.

Environment

MemMachine version 0.3.2

Additional context

No response

When I reproduce this issue in local lab environment, I get another similar errors. This means we have issue to parse Qwen3.5 response.

2026-03-26 06:00:30,508 [INFO] httpx - HTTP Request: POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions "HTTP/1.1 200 OK"
2026-03-26 06:00:30,511 [ERROR] memmachine_server.semantic_memory.semantic_ingestion - Failed to process message 46 for semantic type profile
Traceback (most recent call last):
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/semantic_memory/semantic_ingestion.py", line 163, in process_semantic_type
    commands = await llm_feature_update(
               ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/pydantic/_internal/_validate_call.py", line 34, in wrapper_function
    return await wrapper(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/semantic_memory/semantic_llm.py", line 88, in llm_feature_update
    parsed_output = await model.generate_parsed_response(
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/common/language_model/openai_chat_completions_language_model.py", line 130, in generate_parsed_response
    response = await self._client.with_options(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/resources/chat/completions/completions.py", line 1668, in parse
    return await self._post(
           ^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1884, in post
    return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1674, in request
    return await self._process_response(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1763, in _process_response
    return await api_response.parse()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_response.py", line 432, in parse
    parsed = self._options.post_parser(parsed)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/resources/chat/completions/completions.py", line 1662, in parser
    return _parse_chat_completion(
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_completions.py", line 146, in parse_chat_completion
    "parsed": maybe_parse_content(
              ^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_completions.py", line 196, in maybe_parse_content
    return _parse_content(response_format, message.content)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_completions.py", line 245, in _parse_content
    return cast(ResponseFormatT, model_parse_json(response_format, content))
                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_compat.py", line 171, in model_parse_json
    return model.model_validate_json(data)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/pydantic/main.py", line 766, in model_validate_json
    return cls.__pydantic_validator__.validate_json(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
pydantic_core._pydantic_core.ValidationError: 1 validation error for _SemanticFeatureUpdateRes
  Input should be an object [type=model_type, input_value=[], input_type=list]
    For further information visit https://errors.pydantic.dev/2.12/v/model_type
2026-03-26 06:00:36,661 [INFO] httpx - HTTP Request: POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions "HTTP/1.1 200 OK"
2026-03-26 06:00:36,664 [ERROR] memmachine_server.semantic_memory.semantic_ingestion - Failed to process message 48 for semantic type profile
Traceback (most recent call last):
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/semantic_memory/semantic_ingestion.py", line 163, in process_semantic_type
    commands = await llm_feature_update(
               ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/pydantic/_internal/_validate_call.py", line 34, in wrapper_function
    return await wrapper(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/semantic_memory/semantic_llm.py", line 88, in llm_feature_update
    parsed_output = await model.generate_parsed_response(
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/memmachine_server/common/language_model/openai_chat_completions_language_model.py", line 130, in generate_parsed_response
    response = await self._client.with_options(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/resources/chat/completions/completions.py", line 1668, in parse
    return await self._post(
           ^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1884, in post
    return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1674, in request
    return await self._process_response(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_base_client.py", line 1763, in _process_response
    return await api_response.parse()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_response.py", line 432, in parse
    parsed = self._options.post_parser(parsed)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/resources/chat/completions/completions.py", line 1662, in parser
    return _parse_chat_completion(
           ^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_completions.py", line 146, in parse_chat_completion
    "parsed": maybe_parse_content(
              ^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_completions.py", line 196, in maybe_parse_content
    return _parse_content(response_format, message.content)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/lib/_parsing/_completions.py", line 245, in _parse_content
    return cast(ResponseFormatT, model_parse_json(response_format, content))
                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/openai/_compat.py", line 171, in model_parse_json
    return model.model_validate_json(data)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.12/site-packages/pydantic/main.py", line 766, in model_validate_json
    return cls.__pydantic_validator__.validate_json(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
pydantic_core._pydantic_core.ValidationError: 1 validation error for _SemanticFeatureUpdateRes
  Input should be an object [type=model_type, input_value=[], input_type=list]
    For further information visit https://errors.pydantic.dev/2.12/v/model_type

The root cause is that Qwen3.5 models with thinking mode enabled prepend a <think>...</think> block to their output before the actual JSON, so when generate_parsed_response in openai_responses_language_model.py passes the raw response content to Pydantic's model_validate_json(), the string is no longer valid JSON and the parse fails. The fix would need to strip the <think>...</think> section from the response content before attempting JSON validation — something like extracting only the text after the closing </think> tag, or using a regex to remove the thinking block entirely. This would need to be handled in generate_parsed_response (around line 157) or as a post-processing step on the raw response string before it reaches Pydantic.

Navigation Menu

Repository navigation

[Bug]: Search memory failed with Qwen3.5-xxx thinking mode (default enabled) #1265

Describe the bug

Steps to reproduce

Expected behavior

Environment

Additional context

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[Bug]: Search memory failed with Qwen3.5-xxx thinking mode (default enabled) #1265

Description

Describe the bug

Steps to reproduce

Expected behavior

Environment

Additional context

Activity

pengfeiz901 commented on Mar 26, 2026

JiwaniZakir commented on Apr 3, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions