[BUG] 在 v3.30.0发布之后——Z AI提供的GLM4.6似乎无法正确开启推理，另外 OpenAI Compatible提供的GLM4.6显示推理但未显示推理内容，推理时长始终为0

[CreatCrzay]

Problem (one or two sentences)

Context (who is affected and when)

Z AI提供的GLM4.6似乎无法正确开启推理，另外 OpenAI Compatible提供的GLM4.6显示推理但未显示推理内容，推理时长始终为0

Reproduction steps

/

Expected result

推理正常工作

Actual result

推理未正常工作

Variations tried (optional)

No response

App Version

v3.30.0

API Provider (optional)

None

Model Used (optional)

GLM4.6

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I'm exploring a potential fix for the GLM4.6 reasoning issues with both Z AI and OpenAI Compatible providers. I'll investigate why reasoning isn't working correctly and will open a PR to share progress.

[roomote]

Opened a PR with my attempt at addressing this issue. All CI checks have passed successfully. Feedback welcome!

PR: #9013

The fix adds GLM-4.6-specific reasoning support by detecting when the model is GLM-4.6 and reasoning is enabled, then adding the thinking: { type: "enabled" } parameter to the API request, which is the vendor-specific format required by Z AI's GLM models.

[nokaka]

Yes, the "thinking" of the ZAI model has consistently failed to work in practice. This may have reduced the quality of the output.

[buz-ch]

0.0.7084 now shows thinking traces but as part of the answer, not in a a thinking bar which is probably related to Issue #9041

[CreatCrzay]

0.0.7084 now shows thinking traces but as part of the answer, not in a a thinking bar which is probably related to Issue #9041

在我提供的示例图片中可以看到并无此现象，也许我误解了你的意思？

[buz-ch]

The example is a bit hard for me to understand (translating screenshots is far harder than plain text), so sorry if this is duplicate

[CreatCrzay]

The example is a bit hard for me to understand (translating screenshots is far harder than plain text), so sorry if this is duplicate

Absolutely no need to apologize! (‾▾‾) By the way, I recommend the open-source project "Pot," an excellent screenshot/text translation and OCR tool, especially useful when paired with large AI models.