diff --git a/lib/crewai/src/crewai/llms/providers/openai/completion.py b/lib/crewai/src/crewai/llms/providers/openai/completion.py
index e2d4c3e40..478cec15e 100644
--- a/lib/crewai/src/crewai/llms/providers/openai/completion.py
+++ b/lib/crewai/src/crewai/llms/providers/openai/completion.py
@@ -1929,14 +1929,15 @@ class OpenAICompletion(BaseLLM):
         # request; the first occurrence is recovered in `_call_completions`.
         # Checked after the tools block on purpose: `reasoning_effort` can also
         # arrive through `additional_params`, which bypasses the typed field.
+        # An absent `reasoning_effort` is not safe either: these models apply a
+        # server-side default, so sending nothing is rejected exactly like sending
+        # "high". Only an explicit "none" is accepted alongside tools.
         if (
             params.get("tools")
-            and params.get("reasoning_effort") not in (None, "none")
+            and params.get("reasoning_effort") != "none"
             and not self._supports_reasoning_effort_with_tools(self.model)
         ):
-            requested = params["reasoning_effort"]
-            # Explicit "none" rather than removing the key: gpt-5.6-* apply a
-            # server-side default, so omitting it fails the same way.
+            requested = params.get("reasoning_effort")
             params["reasoning_effort"] = "none"
             logging.debug(
                 'Forcing reasoning_effort="none" (requested %r) for model %r: '
diff --git a/lib/crewai/tests/llms/openai/test_gpt56_tools_reasoning_effort.py b/lib/crewai/tests/llms/openai/test_gpt56_tools_reasoning_effort.py
index 0e25beaa9..b612c87f6 100644
--- a/lib/crewai/tests/llms/openai/test_gpt56_tools_reasoning_effort.py
+++ b/lib/crewai/tests/llms/openai/test_gpt56_tools_reasoning_effort.py
@@ -108,6 +108,20 @@ class TestLearnedFastPath:
         assert params["tools"]
         assert params["tool_choice"] == "auto"
 
+    def test_absent_reasoning_effort_is_also_forced_to_none(self):
+        """Sending nothing is rejected exactly like sending "high".
+
+        These models apply a server-side default, so a learned model with no
+        `reasoning_effort` set would 400 on every call and never benefit from the
+        cache.
+        """
+        llm = build("gpt-5.6-sol")
+        llm._remember_reasoning_effort_conflict()
+
+        params = llm._prepare_completion_params(MESSAGES, tools=TOOLS)
+
+        assert params["reasoning_effort"] == "none"
+
     def test_learning_applies_to_the_additional_params_leak(self):
         """additional_params bypasses the typed field, so it must be checked too."""
         llm = build("gpt-5.5", additional_params={"reasoning_effort": "medium"})