GLM4.6 开启forcereasoning 后，token usage一直为 0

[Runarry]

GLM4.6 开启forcereasoning 后，token usage一直为 0无法在CC中统计到token消耗，开启前是有数据的。

[kierr]

I have the same issue. In the latest versions (2.0.47 onward) I also face an issue where the agent, as its final response before stopping, skips addressing the main focus of the conversation and instead replies to the system message, promising to use its reasoning skills. This didn't happen before. Maybe the system-reminder can be tweaked, following Anthropic's patterns for system-reminder prompting (i.e. forbidden to mention reasoning to the user)

[Runarry]

I created a custom transformer based on forcereasoning and temporarily worked around the issue. Of course, since I’m not very familiar with JavaScript, this code was also completed with the help of Claude Code.

glm-thinking.transformer.js