OpenAI-API-compatible plugin Qwen3 model does not take effect when the "enable_thinking" option is turned off #951

Closed
opened 2026-02-16 10:21:06 -05:00 by yindo · 1 comment
Owner

Originally created by @junjiem on GitHub (Jan 27, 2026).

Self Checks

  • This is only for bug report, if you would like to ask a question, please head to Discussions.
  • I have searched for existing issues Dify issues & Dify Official Plugins, including closed ones.
  • I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
  • [FOR CHINESE USERS] 请务必使用英文提交 Issue,否则会被关闭。谢谢!:)
  • Please do not modify this template :) and fill in all the required fields.

Dify version

1.11.4

Plugin version

0.0.30, 0.0.31

Cloud or Self Hosted

Cloud, Self Hosted (Docker), Self Hosted (Source)

Steps to reproduce

The OpenAI-API-compatible plugin Qwen3 model does not take effect when the "enable_thinking" option is turned off; it still outputs the thinking content.

The "enable_thinking" parameter does not necessarily have to be included in the "chat_template_kwargs" parameter; it can be placed at the same level as other parameters.

✔️ Error log

No response

Originally created by @junjiem on GitHub (Jan 27, 2026). ### Self Checks - [x] This is only for bug report, if you would like to ask a question, please head to [Discussions](https://github.com/langgenius/dify/discussions/categories/general). - [x] I have searched for existing issues [Dify issues](https://github.com/langgenius/dify/issues) & [Dify Official Plugins](https://github.com/langgenius/dify-official-plugins/issues), including closed ones. - [x] I confirm that I am using English to submit this report (我已阅读并同意 [Language Policy](https://github.com/langgenius/dify/issues/1542)). - [x] [FOR CHINESE USERS] 请务必使用英文提交 Issue,否则会被关闭。谢谢!:) - [x] Please do not modify this template :) and fill in all the required fields. ### Dify version 1.11.4 ### Plugin version 0.0.30, 0.0.31 ### Cloud or Self Hosted Cloud, Self Hosted (Docker), Self Hosted (Source) ### Steps to reproduce The OpenAI-API-compatible plugin Qwen3 model does not take effect when the "enable_thinking" option is turned off; it still outputs the thinking content. The "enable_thinking" parameter does not necessarily have to be included in the "chat_template_kwargs" parameter; it can be placed at the same level as other parameters. ### ✔️ Error log _No response_
yindo added the bug label 2026-02-16 10:21:06 -05:00
yindo closed this issue 2026-02-16 10:21:06 -05:00
Author
Owner

@dosubot[bot] commented on GitHub (Jan 27, 2026):

This is a known regression in plugin versions 0.0.28–0.0.31 for the Qwen3 model: disabling enable_thinking does not fully suppress thinking content, and you may still see <think></think> tags or empty reasoning blocks in the output. This happens because the plugin streams a reasoning_content field with empty or newline content, which Dify renders as thinking blocks, regardless of the parameter's placement (inside or outside chat_template_kwargs)—the code sets both for compatibility, but not all backends honor them the same way [source] [source] [source].

Workarounds: filter out <think> tags or empty reasoning blocks on the frontend, or patch the backend/plugin to suppress empty reasoning content before it reaches the UI. Setting reasoning_format to separated in your workflow node may help, but does not fully resolve the issue in these plugin versions. Recent plugin updates (see PR #2151) add support for thinking and thinking_mode parameters alongside enable_thinking, which may improve compatibility for some models, but robust suppression still depends on backend support [source].

For best results, check your model/provider documentation for the correct parameter(s) to use, and consider upgrading when a fix is released.

To reply, just mention @dosu.


How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other

Open in Cursor Join Discord Share on X

@dosubot[bot] commented on GitHub (Jan 27, 2026): <!-- Answer --> This is a known regression in plugin versions 0.0.28–0.0.31 for the Qwen3 model: disabling `enable_thinking` does not fully suppress thinking content, and you may still see `<think></think>` tags or empty reasoning blocks in the output. This happens because the plugin streams a `reasoning_content` field with empty or newline content, which Dify renders as thinking blocks, regardless of the parameter's placement (inside or outside `chat_template_kwargs`)—the code sets both for compatibility, but not all backends honor them the same way [[source]](https://github.com/langgenius/dify-official-plugins/issues/2447) [[source]](https://github.com/langgenius/dify-official-plugins/issues/849) [[source]](https://github.com/langgenius/dify-official-plugins/blob/fafbb6b1e8eaa4c838e7cca5d90c92d58f05854f/models/openai_api_compatible/models/llm/llm.py#L107-L266). Workarounds: filter out `<think>` tags or empty reasoning blocks on the frontend, or patch the backend/plugin to suppress empty reasoning content before it reaches the UI. Setting `reasoning_format` to `separated` in your workflow node may help, but does not fully resolve the issue in these plugin versions. Recent plugin updates (see PR #2151) add support for `thinking` and `thinking_mode` parameters alongside `enable_thinking`, which may improve compatibility for some models, but robust suppression still depends on backend support [[source]](https://github.com/langgenius/dify-official-plugins/issues/2468). For best results, check your model/provider documentation for the correct parameter(s) to use, and consider upgrading when a fix is released. <!-- Dosu Comment Footer --> *To reply, just mention [@dosu](https://go.dosu.dev/dosubot).* --- <sup>How did I do? [Good](https://app.dosu.dev/response-feedback/b36f5bf4-114d-4184-bbf3-600109cafb09?feedback_type=great_response) | [Irrelevant](https://app.dosu.dev/response-feedback/b36f5bf4-114d-4184-bbf3-600109cafb09?feedback_type=irrelevant_answer) | [Incorrect](https://app.dosu.dev/response-feedback/b36f5bf4-114d-4184-bbf3-600109cafb09?feedback_type=incorrect_sources) | [Verbose](https://app.dosu.dev/response-feedback/b36f5bf4-114d-4184-bbf3-600109cafb09?feedback_type=too_verbose) | [Hallucination](https://app.dosu.dev/response-feedback/b36f5bf4-114d-4184-bbf3-600109cafb09?feedback_type=hallucination) | [Report 🐛](https://app.dosu.dev/response-feedback/b36f5bf4-114d-4184-bbf3-600109cafb09?feedback_type=bug_report) | [Other](https://app.dosu.dev/response-feedback/b36f5bf4-114d-4184-bbf3-600109cafb09?feedback_type=other)</sup> [![Open in Cursor](https://dosu.dev/dosu-in-cursor.svg)](https://cursor.com/link/prompt?text=This%20is%20a%20known%20regression%20in%20plugin%20versions%200.0.28%E2%80%930.0.31%20for%20the%20Qwen3%20model%3A%20disabling%20%60enable_thinking%60%20does%20not%20fully%20suppress%20thinking%20content%2C%20and%20you%20may%20still%20see%20%60%3Cthink%3E%3C/think%3E%60%20tags%20or%20empty%20reasoning%20blocks%20in%20the%20output.%20This%20happens%20because%20the%20plugin%20streams%20a%20%60reasoning_content%60%20field%20with%20empty%20or%20newline%20content%2C%20which%20Dify%20renders%20as%20thinking%20blocks%2C%20regardless%20of%20the%20parameter%27s%20placement%20%28inside%20or%20outside%20%60chat_template_kwargs%60%29%E2%80%94the%20code%20sets%20both%20for%20compatibility%2C%20but%20not%20all%20backends%20honor%20them%20the%20same%20way%20%5B%5Bsource%5D%5D%28https%3A//github.com/langgenius/dify-official-plugins/issues/2447%29%20%5B%5Bsource%5D%5D%28https%3A//github.com/langgenius/dify-official-plugins/issues/849%29%20%5B%5Bsource%5D%5D%28https%3A//github.com/langgenius/dify-official-plugins/blob/fafbb6b1e8eaa4c838e7cca5d90c92d58f05854f/models/openai_api_compatible/models/llm/llm.py%23L107-L266%29.%0A%0AWorkarounds%3A%20filter%20out%20%60%3Cthink%3E%60%20tags%20or%20empty%20reasoning%20blocks%20on%20the%20frontend%2C%20or%20patch%20the%20backend/plugin%20to%20suppress%20empty%20reasoning%20content%20before%20it%20reaches%20the%20UI.%20Setting%20%60reasoning_format%60%20to%20%60separated%60%20in%20your%20workflow%20node%20may%20help%2C%20but%20does%20not%20fully%20resolve%20the%20issue%20in%20these%20plugin%20versions.%20Recent%20plugin%20updates%20%28see%20PR%20%232151%29%20add%20support%20for%20%60thinking%60%20and%20%60thinking_mode%60%20parameters%20alongside%20%60enable_thinking%60%2C%20which%20may%20improve%20compatibility%20for%20some%20models%2C%20but%20robust%20suppression%20still%20depends%20on%20backend%20support%20%5B%5Bsource%5D%5D%28https%3A//github.com/langgenius/dify-official-plugins/issues/2468%29.%0A%0AFor%20best%20results%2C%20check%20your%20model/provider%20documentation%20for%20the%20correct%20parameter%28s%29%20to%20use%2C%20and%20consider%20upgrading%20when%20a%20fix%20is%20released.)&nbsp;[![Join Discord](https://img.shields.io/badge/join-5865F2?logo=discord&logoColor=white&label=)](https://go.dosu.dev/discord-bot)&nbsp;[![Share on X](https://img.shields.io/badge/X-share-black)](https://twitter.com/intent/tweet?text=%40dosu_ai%20helped%20me%20solve%20this%20issue!&url=https%3A//github.com/langgenius/dify-official-plugins/issues/2495)
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: langgenius/dify-official-plugins#951