fix: add GLM-4.6 thinking token support for OpenAI-compatible endpoints

[roomote]

This PR attempts to address Issue #8547 by enabling thinking tokens for GLM-4.6 when using OpenAI-compatible custom endpoints.

Problem

GLM-4.6 was not generating thinking tokens when used through OpenAI-compatible custom endpoints because the required "thinking": {"type": "enabled"} parameter was missing from requests.

Solution

• Added detection for GLM-4.6 model variants (glm-4.6, GLM-4.6, glm-4-6, GLM-4-6)
• Include the thinking parameter in requests when GLM-4.6 is detected
• Parse thinking tokens using XmlMatcher for <think> tags
• Handle reasoning_content field in streaming responses
• Added comprehensive tests to verify functionality

Testing

• All existing tests pass ✅
• New tests added for GLM-4.6 functionality ✅
• Tested model detection, parameter addition, and token parsing

Impact

This change only affects GLM-4.6 models when used through OpenAI-compatible endpoints. Other models and providers remain unaffected.

Fixes #8547

Feedback and guidance are welcome!

---

Important

Adds GLM-4.6 thinking token support for OpenAI-compatible endpoints by detecting models, adding parameters, and parsing tokens.

• Behavior:
  • Detects GLM-4.6 model variants (glm-4.6, GLM-4.6, glm-4-6, GLM-4-6) in base-openai-compatible-provider.ts.
  • Adds thinking: { type: "enabled" } parameter for GLM-4.6 models in createStream().
  • Parses thinking tokens using XmlMatcher for <think> tags in createMessage().
  • Handles reasoning_content field in streaming responses in createMessage().
• Testing:
  • Adds tests in base-openai-compatible-provider.spec.ts for model detection, thinking parameter addition, and token parsing.
  • Verifies handling of reasoning_content in responses.
• Impact:
  • Affects only GLM-4.6 models when used through OpenAI-compatible endpoints.

^{This description was created by for aada7cc. You can customize this summary. It will automatically update as commits are pushed.}

[ellipsis-dev]

The isGLM46Model check has a redundant condition (lowerModel === 'glm-4.6') since includes('glm-4.6') already covers it. Consider removing the extra check.

Suggested change

	return lowerModel.includes("glm-4.6") || lowerModel.includes("glm-4-6") || lowerModel === "glm-4.6"
	return lowerModel.includes("glm-4.6") || lowerModel.includes("glm-4-6")

[roomote]

Self-review mode: evaluating my own changes with all the empathy of a linter in a cold data center.

[roomote]

P2: Potential double-emission of reasoning tokens. For GLM-4.6 you already parse via XmlMatcher; if the provider also populates reasoning_content in the same stream, this emits the same reasoning twice. Gate this branch when GLM-4.6 is active or dedupe.

Suggested change

	if (delta && "reasoning_content" in delta && delta.reasoning_content) {
	// Handle reasoning_content if present (avoid double-emitting when GLM '<think>' is parsed)
	if (!isGLM46 && delta && "reasoning_content" in delta && delta.reasoning_content) {
	yield {
	type: "reasoning",
	text: (delta.reasoning_content as string | undefined) || "",
	}
	}

[roomote]

P3: Minor simplification. The final equality check is redundant because .includes("glm-4.6") already covers it. A concise version improves readability.

Suggested change

	return lowerModel.includes("glm-4.6") || lowerModel.includes("glm-4-6") || lowerModel === "glm-4.6"
	protected isGLM46Model(modelId: string): boolean {
	const lowerModel = modelId.toLowerCase()
	return lowerModel.includes("glm-4.6") || lowerModel.includes("glm-4-6")
	}

[roomote]

P3: Unused import ModelInfo can trigger no-unused-vars in stricter configs.

Suggested change

import type { ModelInfo } from "@roo-code/types"

[ChicoPinto70]

Hi, Guys. I tried this PR with ik_llama.cpp running locally the Ubergarm's GLM-4.6 model with Unsloth's chat template (--jinja --chat-template-file) but I still don't have reasoning in the Roo Code. The reasoning works fine in Roo Code with DeepSeek v3.1 running locally and I also get GLM-4.6 reasoning in the ik_llama.cpp built-in webui and in the continue vscode extension. Do I need to do something else in order to make it works with this PR?

I'm using the openai compatible endpoint in Roo Code with "glm-4.6" alias in the ik_llama.cpp