feat: add effort dropdown for claude 4.5 opus (backend) (#6351)

* feat: add effort support (backend)

* fix test_agent_state_schema_unchanged

---------

Co-authored-by: Ari Webb <ari@letta.com>

letta/schemas/llm_config.py

@@ -441,6 +441,9 @@ class LLMConfig(BaseModel):
                 config.put_inner_thoughts_in_kwargs = False
                 if config.enable_reasoner and config.max_reasoning_tokens == 0:
                     config.max_reasoning_tokens = 1024
+                # Set default effort level for Claude Opus 4.5
+                if config.model.startswith("claude-opus-4-5") and config.effort is None:
+                    config.effort = "medium"
                 return config
 
             # Google Gemini 2.5 Pro and Gemini 3: not possible to disable
@@ -495,6 +498,9 @@ class LLMConfig(BaseModel):
                 config.put_inner_thoughts_in_kwargs = False
                 if config.max_reasoning_tokens == 0:
                     config.max_reasoning_tokens = 1024
+                # Set default effort level for Claude Opus 4.5
+                if config.model.startswith("claude-opus-4-5") and config.effort is None:
+                    config.effort = "medium"
             elif cls.is_google_vertex_reasoning_model(config) or cls.is_google_ai_reasoning_model(config):
                 # Handle as non-reasoner until we support summary
                 config.put_inner_thoughts_in_kwargs = True

letta/schemas/model.py

@@ -112,6 +112,7 @@ class Model(LLMConfig, ModelBase):
             enable_reasoner=llm_config.enable_reasoner,
             reasoning_effort=llm_config.reasoning_effort,
             max_reasoning_tokens=llm_config.max_reasoning_tokens,
+            effort=llm_config.effort,
             frequency_penalty=llm_config.frequency_penalty,
             compatibility_type=llm_config.compatibility_type,
             verbosity=llm_config.verbosity,

tests/managers/test_agent_manager.py

@@ -1541,6 +1541,7 @@ async def test_agent_state_schema_unchanged(server: SyncServer):
         "max_tokens",
         "enable_reasoner",
         "reasoning_effort",
+        "effort",
         "max_reasoning_tokens",
         "frequency_penalty",
         "compatibility_type",