feat(llm): GLM dual-region split + catalog refresh

Zhipu serves GLM under two brands with separate accounts (Z.AI
international vs BigModel China); the CLI URL pointed at one while
the openai_client default pointed at the other. Split into glm +
glm-cn with secondary region prompt (same UX as Qwen + MiniMax).
Catalog adds glm-5-turbo and glm-4.5-air per docs.z.ai.

tradingagents/llm_clients/factory.py

@@ -4,7 +4,9 @@ from .base_client import BaseLLMClient
 
 # Providers that use the OpenAI-compatible chat completions API
 _OPENAI_COMPATIBLE = (
-    "openai", "xai", "deepseek", "qwen", "glm",
+    "openai", "xai", "deepseek",
+    "qwen", "qwen-cn",
+    "glm", "glm-cn",
     "minimax", "minimax-cn",
     "ollama", "openrouter",
 )

tradingagents/llm_clients/model_catalog.py

@@ -8,6 +8,25 @@ ModelOption = Tuple[str, str]
 ProviderModeOptions = Dict[str, Dict[str, List[ModelOption]]]
 
 
+# Shared model list for GLM via Z.AI (international) and BigModel (China).
+# Source: docs.z.ai (GLM Coding Plan supported models + LLM guides).
+# All GLM 4.7+ entries support thinking mode via thinking={"type":"enabled"}.
+_GLM_MODELS: Dict[str, List[ModelOption]] = {
+    "quick": [
+        ("GLM-5-Turbo - Fast, switchable thinking modes", "glm-5-turbo"),
+        ("GLM-4.7 - Previous-gen flagship", "glm-4.7"),
+        ("GLM-4.5-Air - Lightweight, cost-efficient", "glm-4.5-air"),
+        ("Custom model ID", "custom"),
+    ],
+    "deep": [
+        ("GLM-5.1 - Latest flagship, 204K ctx", "glm-5.1"),
+        ("GLM-5 - Flagship, 204K ctx", "glm-5"),
+        ("GLM-4.7 - Previous-gen flagship", "glm-4.7"),
+        ("Custom model ID", "custom"),
+    ],
+}
+
+
 # Shared model list for Qwen's global (dashscope-intl) and CN (dashscope) endpoints.
 # Source: modelstudio.console.alibabacloud.com (Featured Models — Flagship + Cost-optimized).
 #
@@ -126,18 +145,10 @@ MODEL_OPTIONS: ProviderModeOptions = {
     # (dashscope) endpoints, so the two provider keys share one model list.
     "qwen": _QWEN_MODELS,
     "qwen-cn": _QWEN_MODELS,
-    "glm": {
-        "quick": [
-            ("GLM-4.7", "glm-4.7"),
-            ("GLM-5", "glm-5"),
-            ("Custom model ID", "custom"),
-        ],
-        "deep": [
-            ("GLM-5.1", "glm-5.1"),
-            ("GLM-5", "glm-5"),
-            ("Custom model ID", "custom"),
-        ],
-    },
+    # GLM: Z.AI (international) and BigModel (China) host the same model
+    # IDs; the two provider keys share one model list.
+    "glm": _GLM_MODELS,
+    "glm-cn": _GLM_MODELS,
     # MiniMax: same model IDs across global (.io) and China (.com) regions,
     # so the two provider keys share one model list.
     "minimax": _MINIMAX_MODELS,

tradingagents/llm_clients/openai_client.py

@@ -139,8 +139,16 @@ _PASSTHROUGH_KWARGS = (
 _PROVIDER_CONFIG = {
     "xai": ("https://api.x.ai/v1", "XAI_API_KEY"),
     "deepseek": ("https://api.deepseek.com", "DEEPSEEK_API_KEY"),
+    # DashScope exposes two regional endpoints with separate accounts; an
+    # international key won't authenticate against the China endpoint and
+    # vice versa (fixes issue #758).
     "qwen": ("https://dashscope-intl.aliyuncs.com/compatible-mode/v1", "DASHSCOPE_API_KEY"),
+    "qwen-cn": ("https://dashscope.aliyuncs.com/compatible-mode/v1", "DASHSCOPE_CN_API_KEY"),
+    # Zhipu exposes the same GLM models under two brands with separate
+    # accounts: Z.AI (international, api.z.ai) and BigModel
+    # (open.bigmodel.cn, China). Keys aren't interchangeable across them.
     "glm": ("https://api.z.ai/api/paas/v4/", "ZHIPU_API_KEY"),
+    "glm-cn": ("https://open.bigmodel.cn/api/paas/v4/", "ZHIPU_CN_API_KEY"),
     # MiniMax exposes two regional endpoints with separate keys; mainland
     # Chinese users hit .com while global users hit .io.
     "minimax": ("https://api.minimax.io/v1", "MINIMAX_API_KEY"),