feat: 模型获取支持 CLI 端点回退，同族优先 chat 后降级 cli

src/services/model/upstream_fetcher.py

@@ -20,8 +20,12 @@ from src.utils.ssl_utils import get_ssl_context
 # 并发请求限制
 MAX_CONCURRENT_REQUESTS = 5
 
-# 只对这些基础 endpoint signature 获取模型列表，CLI 使用相同的上游 API
-MODEL_FETCH_FORMATS = ["openai:chat", "claude:chat", "gemini:chat"]
+# 模型获取格式优先级：同族内优先使用 chat 端点，若无则回退到 cli 端点
+MODEL_FETCH_FORMAT_PRIORITY: list[tuple[str, ...]] = [
+    ("openai:chat", "openai:cli"),
+    ("claude:chat", "claude:cli"),
+    ("gemini:chat", "gemini:cli"),
+]
 
 # Return tuple signature:
 # (models, errors, has_success, upstream_metadata)
@@ -163,17 +167,21 @@ def build_all_format_configs(
     if not format_to_endpoint:
         return []
 
-    # 只对基础 API 格式获取模型，CLI 格式使用相同的上游 API
-    return [
-        {
-            "api_key": api_key_value,
-            "base_url": ep.base_url,
-            "api_format": fmt,
-            "extra_headers": get_extra_headers_from_endpoint(ep),
-        }
-        for fmt in MODEL_FETCH_FORMATS
-        if (ep := format_to_endpoint.get(fmt)) is not None
-    ]
+    # 同族内优先使用 chat 端点，若无则回退到 cli 端点
+    configs: list[dict] = []
+    for candidates in MODEL_FETCH_FORMAT_PRIORITY:
+        fmt = next((f for f in candidates if f in format_to_endpoint), None)
+        if fmt is not None:
+            ep = format_to_endpoint[fmt]
+            configs.append(
+                {
+                    "api_key": api_key_value,
+                    "base_url": ep.base_url,
+                    "api_format": fmt,
+                    "extra_headers": get_extra_headers_from_endpoint(ep),
+                }
+            )
+    return configs
 
 
 async def fetch_models_from_endpoints(

tests/services/test_build_all_format_configs.py

@@ -47,15 +47,46 @@ def test_only_configured_formats_are_included() -> None:
     assert configs[0]["base_url"] == "https://betterclau.de/claude/api.freekey.site"
 
 
-def test_cli_format_only_is_skipped() -> None:
-    """如果只配置了 CLI 格式（不在 MODEL_FETCH_FORMATS 中），应返回空列表。"""
+def test_cli_format_only_fallback() -> None:
+    """只配置了 CLI 格式时，应回退使用 CLI 端点获取模型。"""
     format_to_endpoint = {
         "openai:cli": _make_endpoint("https://api.openai.example.com"),
     }
 
     configs = build_all_format_configs("sk-test-key", format_to_endpoint)  # type: ignore[arg-type]
 
-    assert configs == []
+    assert len(configs) == 1
+    assert configs[0]["api_format"] == "openai:cli"
+    assert configs[0]["base_url"] == "https://api.openai.example.com"
+
+
+def test_chat_takes_priority_over_cli() -> None:
+    """同族同时存在 chat 和 cli 端点时，应优先使用 chat 端点。"""
+    format_to_endpoint = {
+        "openai:chat": _make_endpoint("https://chat.openai.example.com"),
+        "openai:cli": _make_endpoint("https://cli.openai.example.com"),
+    }
+
+    configs = build_all_format_configs("sk-test-key", format_to_endpoint)  # type: ignore[arg-type]
+
+    assert len(configs) == 1
+    assert configs[0]["api_format"] == "openai:chat"
+    assert configs[0]["base_url"] == "https://chat.openai.example.com"
+
+
+def test_mixed_chat_and_cli_fallback() -> None:
+    """一个族有 chat、另一个族只有 cli，应各自正确选择。"""
+    format_to_endpoint = {
+        "openai:chat": _make_endpoint("https://chat.openai.example.com"),
+        "claude:cli": _make_endpoint("https://cli.claude.example.com"),
+    }
+
+    configs = build_all_format_configs("sk-test-key", format_to_endpoint)  # type: ignore[arg-type]
+
+    assert len(configs) == 2
+    by_fmt = {c["api_format"]: c for c in configs}
+    assert by_fmt["openai:chat"]["base_url"] == "https://chat.openai.example.com"
+    assert by_fmt["claude:cli"]["base_url"] == "https://cli.claude.example.com"
 
 
 def test_empty_format_to_endpoint() -> None: