GLM 4.7 thinking process not properly formatted - missing leading tags

[ukyonagata0105]

Description

When using GLM 4.7 model, the thinking process is not being properly formatted. The opening tag (e.g., <think>) at the beginning of the thinking content is missing, causing all reasoning output to appear as plain text in the UI.

Expected Behavior

The thinking process should be wrapped with proper tags like:

<think>
[reasoning content]
</think>

This allows the UI to properly hide/display the thinking content.

Actual Behavior

Only the closing tag is present, or the entire reasoning appears as plain text without any formatting. The opening <think> tag is missing from the beginning of the thinking process.

Steps to Reproduce

1. Configure opencode to use GLM 4.7 model (Zhipu AI)
2. Ask a question that triggers tool usage or complex reasoning
3. Observe the output - the thinking process appears as plain text instead of being properly formatted

Environment

• Model: Zhipu GLM-4.7
• Provider: Custom/Local LLM provider using vLLM or similar

Additional Context

This may be related to how the streaming response is parsed for GLM 4.7, specifically in the JSON parsing logic that handles reasoning_content fields.

Plugins

No response

OpenCode version

1.1.13

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [Bug] JSON Parse Error with Zhipu GLM-4.7: Stream chunks are concatenated incorrectly #7692: JSON Parse Error with Zhipu GLM-4.7 - stream chunks are concatenated incorrectly, which could cause parsing issues with thinking tags
• GLM 4.7 on Zai coding plan puts tool calls inside the thinking/reasoning tag. #6708: GLM 4.7 on Zai coding plan puts tool calls inside the thinking/reasoning tag - related issue with thinking tag handling
• Error: Invalid signature in thinking block #6418: Invalid signature in thinking block - thinking tag format errors when switching models from GLM 4.7
• Error "reasoning part reasoning-0 not found" from tool calls using GLM 4.5 Air #6486: Reasoning part not found error from GLM tool calls - similar thinking/reasoning content parsing issues

Feel free to ignore if none of these address your specific case.

[goniz]

Hi, this issue is related to how you serve your local LLM.
Your inference server (llama.cpp? Vllm?) needs to be correctly configured with the prompt template and to serve the correct fields of reasoning_content etc (depends on the API, completions/responses)

[pawanjenu]

facing the same issue. I use VLLM