feat: revert model_settings (#6089)

tests/configs/llm_model_configs/gemini-2.5-flash.json

@@ -6,5 +6,5 @@
   "model_wrapper": null,
   "put_inner_thoughts_in_kwargs": true,
   "enable_reasoner": true,
-  "max_reasoning_tokens": 1000
+  "max_reasoning_tokens": 20000
 }

tests/managers/test_agent_manager.py

@@ -1250,7 +1250,7 @@ async def test_agent_state_schema_unchanged(server: SyncServer):
     from letta.schemas.group import Group
     from letta.schemas.llm_config import LLMConfig
     from letta.schemas.memory import Memory
-    from letta.schemas.model import ModelSettings
+    from letta.schemas.model import EmbeddingModelSettings, ModelSettings
     from letta.schemas.response_format import ResponseFormatUnion
     from letta.schemas.source import Source
     from letta.schemas.tool import Tool
@@ -1271,10 +1271,9 @@ async def test_agent_state_schema_unchanged(server: SyncServer):
         "agent_type": AgentType,
         # LLM information
         "llm_config": LLMConfig,
-        "model": str,
-        "embedding": str,
+        "model": ModelSettings,
+        "embedding": EmbeddingModelSettings,
         "embedding_config": EmbeddingConfig,
-        "model_settings": ModelSettings,
         "response_format": (ResponseFormatUnion, type(None)),
         # State fields
         "description": (str, type(None)),

tests/sdk_v1/agents_test.py

@@ -5,9 +5,11 @@ AGENTS_CREATE_PARAMS = [
         "caren_agent",
         {"name": "caren", "model": "openai/gpt-4o-mini", "embedding": "openai/text-embedding-3-small"},
         {
-            # Verify model_settings is populated with config values
-            # Note: The 'model' field itself is separate from model_settings
-            "model_settings": {"max_output_tokens": 4096, "parallel_tool_calls": False}
+            # Verify model field contains the model name and settings
+            # Note: we override 'model' here since the input is a string but the output is a ModelSettings object
+            "model": {"model": "gpt-4o-mini", "max_output_tokens": 4096, "parallel_tool_calls": False},
+            # Note: we override 'embedding' here since it's currently not populated in AgentState (remains None)
+            "embedding": None,
         },
         None,
     ),
@@ -18,8 +20,8 @@ AGENTS_MODIFY_PARAMS = [
         "caren_agent",
         {"name": "caren_updated"},
         {
-            # After modifying just the name, model_settings should still be present
-            "model_settings": {"max_output_tokens": 4096, "parallel_tool_calls": False}
+            # After modifying just the name, model field should still be present and unchanged
+            "model": {"model": "gpt-4o-mini", "max_output_tokens": 4096, "parallel_tool_calls": False}
         },
         None,
     ),