fix(providers): add Qwen model-level thinking style mapping (#5023)

Add _QWEN_THINKING_MODELS to _MODEL_THINKING_STYLES with enable_thinking style. Prevents Qwen 3.5/3.6/3.7 models from exposing raw reasoning content in chat responses. Closes #4934

nanobot/providers/openai_compat_provider.py

@@ -99,9 +99,20 @@ _THINKING_STYLE_MAP: dict[str, Any] = {
 _GATEWAY_REASONING_STYLE_MAP: dict[str, Any] = {
     "reasoning_effort": lambda effort: {"reasoning": {"effort": effort}},
 }
+_QWEN_THINKING_MODELS: frozenset[str] = frozenset({
+    "qwen3.7-max",
+    "qwen3.7-plus",
+    "qwen3.6-max-preview",
+    "qwen3.6-plus",
+    "qwen3.6-flash",
+    "qwen3.5-plus",
+    "qwen3.5-flash",
+})
+
 _MODEL_THINKING_STYLES: dict[str, str] = {
     **dict.fromkeys(_KIMI_THINKING_MODELS, "thinking_type"),
     **dict.fromkeys(_MIMO_THINKING_MODELS, "thinking_type"),
+    **dict.fromkeys(_QWEN_THINKING_MODELS, "enable_thinking"),
 }
 
 

tests/providers/test_litellm_kwargs.py

@@ -1707,6 +1707,28 @@ def test_dashscope_thinking_disabled_for_none_string() -> None:
     assert "reasoning_effort" not in kw
 
 
+def test_qwen_thinking_enabled_via_model_level_mapping() -> None:
+    """Non-DashScope providers (e.g. OpenRouter) must pick up model-level
+    enable_thinking for Qwen models when reasoning_effort is set."""
+    kw = _build_kwargs_for("openrouter", "qwen/qwen3.6-flash", reasoning_effort="medium")
+    assert kw["extra_body"] == {"enable_thinking": True, "reasoning": {"effort": "medium"}}
+
+
+def test_qwen_thinking_disabled_via_model_level_mapping() -> None:
+    """reasoning_effort='none' must send enable_thinking: False via model-level
+    mapping on non-DashScope providers. OpenRouter also emits its own
+    reasoning.effort alongside the provider-level thinking control."""
+    kw = _build_kwargs_for("openrouter", "qwen/qwen3.5-flash", reasoning_effort="none")
+    assert kw["extra_body"] == {"enable_thinking": False, "reasoning": {"effort": "none"}}
+
+
+def test_qwen_no_extra_body_when_reasoning_effort_omitted() -> None:
+    """Without reasoning_effort the model-level mapping must not inject extra_body
+    on its own — the provider default applies."""
+    kw = _build_kwargs_for("openrouter", "qwen/qwen3.6-flash", reasoning_effort=None)
+    assert "extra_body" not in kw
+
+
 def test_deepseek_no_backfill_when_reasoning_effort_none_string() -> None:
     """reasoning_effort='none' must NOT trigger reasoning_content backfill (thinking inactive)."""
     spec = find_by_name("deepseek")