diff --git a/.env.example b/.env.example
index 4ead17f..3021cda 100644
--- a/.env.example
+++ b/.env.example
@@ -14,7 +14,7 @@ REDIS_URL=redis://localhost:6379/0
 # LLM Provider Configuration
 # Supported providers: "gemini", "openai", "fake"
 LLM_PROVIDER=gemini
-LLM_MODEL=gemini-2.0-flash
+LLM_MODEL=gemini-3.7-flash
 GEMINI_API_KEY=your_gemini_api_key_here
 OPENAI_API_KEY=your_openai_api_key_here
 
diff --git a/agentkit/config.py b/agentkit/config.py
index 1121ada..7ebe44f 100644
--- a/agentkit/config.py
+++ b/agentkit/config.py
@@ -32,7 +32,7 @@ class Settings(BaseSettings):
         description="LLM provider: gemini, openai, or fake.",
     )
     LLM_MODEL: str = Field(
-        default="gemini-2.0-flash",
+        default="gemini-3.7-flash",
         description="LLM model identifier.",
     )
     GEMINI_API_KEY: str | None = Field(
diff --git a/agentkit/llm/factory.py b/agentkit/llm/factory.py
index 56168d5..4de7918 100644
--- a/agentkit/llm/factory.py
+++ b/agentkit/llm/factory.py
@@ -38,7 +38,7 @@ def create_llm_client(
     normalized_provider = provider.strip().lower()
 
     if normalized_provider == "gemini":
-        default_model = model or "gemini-2.0-flash"
+        default_model = model or "gemini-3.7-flash"
         return GeminiClient(api_key=api_key, model=default_model, **kwargs)
 
     if normalized_provider == "openai":
diff --git a/agentkit/llm/gemini.py b/agentkit/llm/gemini.py
index d9815a6..b0dc6fe 100644
--- a/agentkit/llm/gemini.py
+++ b/agentkit/llm/gemini.py
@@ -18,14 +18,14 @@ class GeminiClient(LLMClient):
     def __init__(
         self,
         api_key: str | None = None,
-        model: str = "gemini-2.0-flash",
+        model: str = "gemini-3.7-flash",
         client: Any | None = None,
     ) -> None:
         """Initialize the Gemini client adapter.
 
         Args:
             api_key: Gemini API key. If omitted, google-genai attempts ADC/env lookup.
-            model: Gemini model identifier (default: gemini-2.0-flash).
+            model: Gemini model identifier (default: gemini-3.7-flash).
             client: Optional pre-configured genai.Client instance (useful for mocking).
         """
         self.model = model
@@ -107,7 +107,7 @@ async def chat(
                 )
                 contents.append(
                     types.Content(
-                        role="tool",
+                        role="user",
                         parts=[
                             types.Part.from_function_response(
                                 name=tool_name,
diff --git a/docker-compose.yml b/docker-compose.yml
index 701ba51..d743b0e 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -76,7 +76,7 @@ services:
       # Compose keeps ${VAR:-} so `down`, `ps` and `logs` still work without it.
       - API_KEYS=${API_KEYS:-}
       - LLM_PROVIDER=${LLM_PROVIDER:-gemini}
-      - LLM_MODEL=${LLM_MODEL:-gemini-2.0-flash}
+      - LLM_MODEL=${LLM_MODEL:-gemini-3.7-flash}
       - GEMINI_API_KEY=${GEMINI_API_KEY:-}
       - OPENAI_API_KEY=${OPENAI_API_KEY:-}
     volumes:
diff --git a/tests/unit/test_config.py b/tests/unit/test_config.py
index f06845d..05d4b56 100644
--- a/tests/unit/test_config.py
+++ b/tests/unit/test_config.py
@@ -14,7 +14,7 @@ def test_default_settings() -> None:
         API_KEYS="key1,key2",
     )
     assert settings.LLM_PROVIDER == "gemini"
-    assert settings.LLM_MODEL == "gemini-2.0-flash"
+    assert settings.LLM_MODEL == "gemini-3.7-flash"
     assert settings.MAX_STEPS == 10
     assert settings.RUN_TIMEOUT_S == 60
     assert settings.TOOL_TIMEOUT_S == 15
diff --git a/tests/unit/test_gemini_adapter.py b/tests/unit/test_gemini_adapter.py
index 1dc9d9d..495e650 100644
--- a/tests/unit/test_gemini_adapter.py
+++ b/tests/unit/test_gemini_adapter.py
@@ -126,6 +126,9 @@ async def test_gemini_multi_turn_history_translation(mock_genai_client: MagicMoc
     response = await adapter.chat(messages)
     assert response.text == "Result is 25."
     assert mock_genai_client.aio.models.generate_content.called
+    _, kwargs = mock_genai_client.aio.models.generate_content.call_args
+    contents = kwargs["contents"]
+    assert contents[2].role == "user"
 
 
 @pytest.mark.asyncio
@@ -192,15 +195,15 @@ async def test_gemini_thought_signature_capture_and_forward(mock_genai_client: M
     assert model_part.function_call.name == "calculator"
     assert model_part.thought_signature == b"opaque_thought_sig_xyz"
 
-    # Tool message must have role="tool" and function_response name resolved to "calculator"
+    # Tool message must have role="user" and function_response name resolved to "calculator"
     tool_content = contents[2]
-    assert tool_content.role == "tool"
+    assert tool_content.role == "user"
     tool_part = tool_content.parts[0]
     assert tool_part.function_response.name == "calculator"
     assert tool_part.function_response.response == {"result": "4"}
 
 
 def test_gemini_default_model(mock_genai_client: MagicMock) -> None:
-    """Verify GeminiClient defaults to gemini-2.0-flash."""
+    """Verify GeminiClient defaults to gemini-3.7-flash."""
     adapter = GeminiClient(api_key="fake-key", client=mock_genai_client)
-    assert adapter.model == "gemini-2.0-flash"
+    assert adapter.model == "gemini-3.7-flash"
diff --git a/tests/unit/test_llm_factory.py b/tests/unit/test_llm_factory.py
index 2f4afc9..aacb779 100644
--- a/tests/unit/test_llm_factory.py
+++ b/tests/unit/test_llm_factory.py
@@ -26,6 +26,10 @@ def test_create_gemini_client() -> None:
     assert client.model == "gemini-2.5-pro"
     assert client.api_key == "test-key-gemini"
 
+    client_default = create_llm_client(provider="gemini", api_key="test-key-gemini")
+    assert isinstance(client_default, GeminiClient)
+    assert client_default.model == "gemini-3.7-flash"
+
 
 def test_create_openai_client() -> None:
     client = create_llm_client(