[Feature]: GLM-4.7 thinking mode support

[hi-one-gg]

Prerequisites

• I have searched existing issues and discussions to avoid duplicates
• This feature request is specific to oh-my-opencode (not OpenCode core)
• I have read the documentation

Problem Description

[Request] Support for Z.AI GLM Thinking Mode in Oh My OpenCode

Hello, I am the user who previously inquired about activating Thinking Mode in the GLM-4.7 model.

Proposal: Enable Thinking Mode for Z.AI GLM Models

According to the Official Z.AI API Documentation, GLM models have Thinking Mode enabled by default, but users can explicitly disable it if desired. Notably, unlike other models, it also supports a dedicated Deep Thinking Mode.

Current Issues & Observations

• Inconsistency with ulw: When using the GLM-4.7 model in Oh My OpenCode, applying the ulw command actually hides the thinking process. Conversely, the thinking process is visible only when ulw is not used.
• UI Feedback: There is currently no UI indicator to confirm whether Thinking Mode is active, and the Ctrl+T shortcut does not seem to trigger any response for this model. This makes it confusing to determine the optimal settings.

Suggested Implementation

I suggest the following integration for the GLM-4.7 model:

1. Standard: Set Thinking Mode to high by default.
2. Deep Thinking: When the user reaches the max level (e.g., via Ctrl+T), the API should be called using the Deep Thinking Mode.

---

I recently watched your interview video and really enjoyed it. I am a big fan of this tool and use it frequently. Keep up the amazing work! Fighting!

Proposed Solution

.

Alternatives Considered

No response

Doctor Output (Optional)

Additional Context

No response

Feature Type

New Agent

Contribution

• I'm willing to submit a PR for this feature
• I can help with testing
• I can help with documentation

---

Original (auto-translated)

Problem Description

안녕하세요 GLM-4.7모델에서 Thinking모드 활성화에 대해 질문 드렸던 사람입니다.

오마이오픈코드에서 Z.AI의 GLM모델도 thinking모드 지원을 부탁드립니다.

공식 API 문서에 따르면 GLM 모델은 thinking모드가 기본적으로 켜져있지만 사용자가 원하면 끈채로 api를 호출할 수 있다고 안내하고 있습니다. 그리고 다른 모델들과 다르게 deep thinking모드를 추가로 지원합니다.

GLM4.7 모델을 사용하여 오마이오픈코드를 쓸 때 ulw 을 사용하면 모델의 Thinking과정이 보이지 않습니다. 오히려 ulw 없이 그냥 사용하면 Thinking과정이 보입니다. 따라서 ulw을 써야할지 말아야할지 다소 혼란스럽습니다. 그 어디에도 Thinking모드 활성화가 됐다는 UI가 보이지 않기 때문입니다. (Crtl+T를 눌러도 반응이 없습니다.)

제 생각엔 GLM4.7모델은 기본적으로 high 단계로 thinking모드가 켜지고 Crtl+t를 눌러 max단계에 도달시 DeepThinking 모드로 API를 호출하면 좋을 것 같습니다.

최근 인터뷰한 영상 잘봤습니다. 그리고 정말 잘 쓰고 있습니다. 화이팅하십시오!

[code-yeongyu]

@sisyphus-dev-ai debug this ulw

[sisyphus-dev-ai]

GLM-4.7 Thinking Mode Support Implemented

I've implemented thinking mode support for GLM-4.7 via the zai-coding-plan provider.

PR Created

#1056

What was added

1. THINKING_CONFIGS - Added zai-coding-plan provider config:

   "zai-coding-plan": {
     providerOptions: {
       "zai-coding-plan": {
         extra_body: {
           thinking: {
             type: "enabled",
             clear_thinking: false,  // Preserved Thinking
           },
         },
       },
     },
   }

2. THINKING_CAPABLE_MODELS - Added GLM pattern detection:

   "zai-coding-plan": ["glm"]

3. Tests - 7 new tests for GLM thinking mode

How it works

When you include "think" keywords in your prompt while using zai-coding-plan/glm-4.7, the think-mode hook will:

1. Detect the keyword
2. Inject the thinking configuration
3. Enable extended reasoning with clear_thinking: false (Preserved Thinking mode)

No model variant change is needed since GLM-4.7 has thinking enabled by default.

Verification

• ✅ All 54 switcher tests pass
• ✅ All 71 think-mode tests pass
• ✅ TypeScript type check passes
• ✅ Build succeeds

[justsisyphus]

Status Update

GLM-4.7 thinking mode support has been implemented and is under review in PR #1056.

What's been done:

✅ Added zai-coding-plan provider to THINKING_CONFIGS with proper extra_body.thinking configuration
✅ Added GLM pattern detection to THINKING_CAPABLE_MODELS
✅ Comprehensive test coverage (7 new tests for GLM models)
✅ All tests passing (54 switcher tests, 71 think-mode tests)
✅ TypeScript type check and build verified

How it works:

When using zai-coding-plan/glm-4.7, the think-mode hook will:

1. Detect "think" keywords in your prompt
2. Inject thinking configuration with clear_thinking: false (preserves reasoning)
3. Enable extended reasoning without requiring model variant changes

Next steps:

• PR feat(think-mode): add GLM-4.7 thinking mode support #1056 is awaiting review and merge to dev branch
• Once merged, GLM-4.7 thinking mode will be available in the next release

The implementation follows Z.AI's API specification for thinking mode support.

[code-yeongyu]

This issue has been automatically translated to English. Please use English for future communications to help our international community. Thank you!

[AJINGOUP]

Small regional/provider nuance: this issue (#1030) is about the Z.AI (international) GLM Coding Plan, so the examples and (later) thinking-mode support naturally target zai-coding-plan/*.

However, for Mainland China users (open.bigmodel.cn / Zhipu Open Platform), the GLM Coding Plan provider commonly used in the OpenCode / oh-my-opencode ecosystem is zhipuai-coding-plan/*, not zai-coding-plan/*. This causes two practical problems:

1. Following the current installer/docs that enable “Z.ai Coding Plan” ends up hardcoding models like zai-coding-plan/glm-4.7 (and in some places Librarian is forced to use it). For mainland subscriptions, users must manually change the model prefix to zhipuai-coding-plan/... to make it work.
2. The thinking-mode integration (PR feat(think-mode): add GLM-4.7 thinking mode support #1056) appears to add configs specifically for zai-coding-plan. If a user is on zhipuai-coding-plan/*, the same thinking toggle logic may not match/apply.

Suggestion (if you want to support both smoothly):

• Make docs/installer explicitly distinguish the two Coding Plans and offer a choice/flag:
  • International: zai-coding-plan/*
  • Mainland: zhipuai-coding-plan/*
• Treat zhipuai-coding-plan as an equivalent provider in the thinking-mode config (same extra_body.thinking handling).
• Document the Coding Plan base URL difference:
  • Z.AI: https://api.z.ai/api/coding/paas/v4
  • Mainland: https://open.bigmodel.cn/api/coding/paas/v4

This would prevent the common “installed successfully but doesn’t run / thinking toggle doesn’t work” confusion for mainland users.