diff --git a/nanobot/providers/openai_compat_provider.py b/nanobot/providers/openai_compat_provider.py
index b49a0b32c..950907212 100644
--- a/nanobot/providers/openai_compat_provider.py
+++ b/nanobot/providers/openai_compat_provider.py
@@ -297,22 +297,23 @@ class OpenAICompatProvider(LLMProvider):
         if reasoning_effort:
             kwargs["reasoning_effort"] = reasoning_effort
 
-        # Provider-specific thinking parameters
-        if spec:
-            # Refer: https://docs.byteplus.com/en/docs/ModelArk/1449737#adjust-reasoning-length
-            # The agent will stop thinking if reasoning_effort is minimal or None. Otherwise, it will think.
-            thinking_enabled = (
-                reasoning_effort is not None and reasoning_effort.lower() != "minimal"
-            )
+        # Provider-specific thinking parameters.
+        # Only sent when reasoning_effort is explicitly configured so that
+        # the provider default is preserved otherwise.
+        if spec and reasoning_effort is not None:
+            thinking_enabled = reasoning_effort.lower() != "minimal"
+            extra: dict[str, Any] | None = None
             if spec.name == "dashscope":
-                # Qwen: extra_body={"enable_thinking": True/False}
-                kwargs["extra_body"] = {"enable_thinking": thinking_enabled}
-            elif spec.name in ("volcengine", "volcengine_coding_plan"):
-                # VolcEngine/Byteplus ModelArk: extra_body={"thinking": {"type": "enabled"/"disabled"}}
-                kwargs["extra_body"] = {
+                extra = {"enable_thinking": thinking_enabled}
+            elif spec.name in (
+                "volcengine", "volcengine_coding_plan",
+                "byteplus", "byteplus_coding_plan",
+            ):
+                extra = {
                     "thinking": {"type": "enabled" if thinking_enabled else "disabled"}
                 }
-
+            if extra:
+                kwargs.setdefault("extra_body", {}).update(extra)
 
         if tools:
             kwargs["tools"] = tools
diff --git a/tests/providers/test_litellm_kwargs.py b/tests/providers/test_litellm_kwargs.py
index 1be505872..2e885e165 100644
--- a/tests/providers/test_litellm_kwargs.py
+++ b/tests/providers/test_litellm_kwargs.py
@@ -307,3 +307,54 @@ async def test_openai_compat_stream_watchdog_returns_error_on_stall(monkeypatch)
     assert result.finish_reason == "error"
     assert result.content is not None
     assert "stream stalled" in result.content
+
+
+# ---------------------------------------------------------------------------
+# Provider-specific thinking parameters (extra_body)
+# ---------------------------------------------------------------------------
+
+def _build_kwargs_for(provider_name: str, model: str, reasoning_effort=None):
+    spec = find_by_name(provider_name)
+    with patch("nanobot.providers.openai_compat_provider.AsyncOpenAI"):
+        p = OpenAICompatProvider(api_key="k", default_model=model, spec=spec)
+    return p._build_kwargs(
+        messages=[{"role": "user", "content": "hi"}],
+        tools=None, model=model, max_tokens=1024, temperature=0.7,
+        reasoning_effort=reasoning_effort, tool_choice=None,
+    )
+
+
+def test_dashscope_thinking_enabled_with_reasoning_effort() -> None:
+    kw = _build_kwargs_for("dashscope", "qwen3-plus", reasoning_effort="medium")
+    assert kw["extra_body"] == {"enable_thinking": True}
+
+
+def test_dashscope_thinking_disabled_for_minimal() -> None:
+    kw = _build_kwargs_for("dashscope", "qwen3-plus", reasoning_effort="minimal")
+    assert kw["extra_body"] == {"enable_thinking": False}
+
+
+def test_dashscope_no_extra_body_when_reasoning_effort_none() -> None:
+    kw = _build_kwargs_for("dashscope", "qwen-turbo", reasoning_effort=None)
+    assert "extra_body" not in kw
+
+
+def test_volcengine_thinking_enabled() -> None:
+    kw = _build_kwargs_for("volcengine", "doubao-seed-2-0-pro", reasoning_effort="high")
+    assert kw["extra_body"] == {"thinking": {"type": "enabled"}}
+
+
+def test_byteplus_thinking_disabled_for_minimal() -> None:
+    kw = _build_kwargs_for("byteplus", "doubao-seed-2-0-pro", reasoning_effort="minimal")
+    assert kw["extra_body"] == {"thinking": {"type": "disabled"}}
+
+
+def test_byteplus_no_extra_body_when_reasoning_effort_none() -> None:
+    kw = _build_kwargs_for("byteplus", "doubao-seed-2-0-pro", reasoning_effort=None)
+    assert "extra_body" not in kw
+
+
+def test_openai_no_thinking_extra_body() -> None:
+    """Non-thinking providers should never get extra_body for thinking."""
+    kw = _build_kwargs_for("openai", "gpt-4o", reasoning_effort="medium")
+    assert "extra_body" not in kw