feat(gemini): expose long decode continuation capability

litellm/__init__.py

@@ -2379,6 +2379,7 @@ if TYPE_CHECKING:
     supports_audio_input: Callable[..., bool]
     supports_video_input: Callable[..., bool]
     supports_box_questions: Callable[..., bool]
+    supports_continuation: Callable[..., bool]
     supports_audio_output: Callable[..., bool]
     supports_system_messages: Callable[..., bool]
     supports_reasoning: Callable[..., bool]

litellm/_lazy_imports_registry.py

@@ -39,6 +39,7 @@ UTILS_NAMES: Final = (
     "supports_audio_input",
     "supports_video_input",
     "supports_box_questions",
+    "supports_continuation",
     "supports_audio_output",
     "supports_system_messages",
     "supports_reasoning",
@@ -497,6 +498,7 @@ _UTILS_IMPORT_MAP: Final = {
     "supports_audio_input": (".utils", "supports_audio_input"),
     "supports_video_input": (".utils", "supports_video_input"),
     "supports_box_questions": (".utils", "supports_box_questions"),
+    "supports_continuation": (".utils", "supports_continuation"),
     "supports_audio_output": (".utils", "supports_audio_output"),
     "supports_system_messages": (".utils", "supports_system_messages"),
     "supports_reasoning": (".utils", "supports_reasoning"),

litellm/model_prices_and_context_window_backup.json

@@ -76646,6 +76646,7 @@
         "supports_audio_input": true,
         "supports_audio_output": false,
         "supports_computer_use": true,
+        "supports_continuation": true,
         "supports_function_calling": true,
         "supports_minimal_reasoning_effort": false,
         "supports_native_streaming": true,
@@ -76705,6 +76706,7 @@
         ],
         "supports_audio_input": true,
         "supports_computer_use": true,
+        "supports_continuation": true,
         "supports_function_calling": true,
         "supports_minimal_reasoning_effort": false,
         "supports_native_streaming": true,

litellm/types/utils.py

@@ -210,6 +210,7 @@ class ProviderSpecificModelInfo(TypedDict, total=False):
     supports_audio_input: bool | None
     supports_video_input: ReadOnly[bool | None]
     supports_box_questions: ReadOnly[bool | None]
+    supports_continuation: ReadOnly[bool | None]
     supports_embedding_image_input: bool | None
     supports_audio_output: bool | None
     supports_pdf_input: bool | None

litellm/utils.py

@@ -3458,6 +3458,11 @@ def is_explicitly_disabled_factory(model: str, custom_llm_provider: str | None,
         return False
 
 
+def supports_continuation(model: str, custom_llm_provider: str | None = None) -> bool:
+    """Check if the model declares support for long-decode continuation."""
+    return supports_factory(model=model, custom_llm_provider=custom_llm_provider, key="supports_continuation")
+
+
 def supports_audio_input(model: str, custom_llm_provider: str | None = None) -> bool:
     """Check if a given model supports audio input in a chat completion call"""
     return supports_factory(model=model, custom_llm_provider=custom_llm_provider, key="supports_audio_input")
@@ -7343,6 +7348,7 @@ def get_model_info_helper(
                 supports_audio_input=_model_info.get("supports_audio_input", None),
                 supports_video_input=_model_info.get("supports_video_input", None),
                 supports_box_questions=_model_info.get("supports_box_questions", None),
+                supports_continuation=_model_info.get("supports_continuation", None),
                 supports_audio_output=_model_info.get("supports_audio_output", None),
                 supports_pdf_input=_model_info.get("supports_pdf_input", None),
                 supports_embedding_image_input=_model_info.get("supports_embedding_image_input", None),
@@ -7505,6 +7511,7 @@ def get_model_info(
             supports_audio_input: Optional[bool]
             supports_video_input: Optional[bool]
             supports_box_questions: Optional[bool]
+            supports_continuation: Optional[bool]
             supports_audio_output: Optional[bool]
             supports_pdf_input: Optional[bool]
             supports_web_search: Optional[bool]

model_prices_and_context_window.json

@@ -76646,6 +76646,7 @@
         "supports_audio_input": true,
         "supports_audio_output": false,
         "supports_computer_use": true,
+        "supports_continuation": true,
         "supports_function_calling": true,
         "supports_minimal_reasoning_effort": false,
         "supports_native_streaming": true,
@@ -76705,6 +76706,7 @@
         ],
         "supports_audio_input": true,
         "supports_computer_use": true,
+        "supports_continuation": true,
         "supports_function_calling": true,
         "supports_minimal_reasoning_effort": false,
         "supports_native_streaming": true,

tests/unit/proxy/test_proxy_server_endpoints_and_startup.py

@@ -6226,6 +6226,35 @@ def test_model_info_v1_list_skips_fastapi_jsonable_encoder(monkeypatch):
     assert encoder_spy.call_count == 0
 
 
+
+@pytest.mark.parametrize("route", ("/model/info", "/v1/model/info"))
+def test_model_info_exposes_continuation_and_deployment_override(
+    route: str, local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    deployments: Final = [
+        {"model_name": "default", "litellm_params": {"model": "gemini/gemini-4-argon"}},
+        {
+            "model_name": "override",
+            "litellm_params": {"model": "vertex_ai/gemini-4-argon"},
+            "model_info": {"supports_continuation": False},
+        },
+    ]
+    router: Final = litellm.Router(model_list=deployments)
+    monkeypatch.setattr(proxy_server_module, "llm_router", router)
+    monkeypatch.setattr(proxy_server_module, "llm_model_list", router.model_list)
+    monkeypatch.setattr(proxy_server_module, "prisma_client", None)
+    monkeypatch.setattr(proxy_server_module, "general_settings", {"infer_model_from_keys": False})
+    monkeypatch.setattr(proxy_server_module, "user_model", None)
+    monkeypatch.setitem(
+        app.dependency_overrides,
+        user_api_key_auth,
+        lambda: UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN, api_key=MASTER_KEY, models=[], team_models=[]),
+    )
+    response: Final = TestClient(app).get(route)
+    assert response.status_code == 200
+    flags: Final = {row["model_name"]: row["model_info"]["supports_continuation"] for row in response.json()["data"]}
+    assert flags == {"default": True, "override": False}
+
 def test_model_info_v1_cli_model_returns_single_deployment_as_json(monkeypatch):
     """
     A proxy started with `litellm --model <name>` answers /model/info with one deployment

tests/unit/test_model_prices_and_context_window.py

@@ -44,6 +44,16 @@ def test_azure_audio_snapshot_capabilities_match_declared_modalities(model: str,
     assert (litellm.supports_audio_input(model), litellm.supports_audio_output(model)) == expected
 
 
+
+@pytest.mark.parametrize("model", ("gemini/gemini-4-argon", "vertex_ai/gemini-4-argon"))
+def test_gemini_long_decode_capability_is_exposed(model: str, local_model_cost_map: None) -> None:
+    # Google long decode documentation, verified 2026-10-30: https://ai.google.dev/gemini-api/docs/generate-content/long-decode
+    catalog: Final = TypeAdapter(dict[str, JsonValue]).validate_json(PRICES_PATH.read_bytes())
+    backup: Final = TypeAdapter(dict[str, JsonValue]).validate_json(BACKUP_PRICES_PATH.read_bytes())
+    assert catalog[model] == backup[model]
+    assert litellm.get_model_info(model)["supports_continuation"] is True
+    assert litellm.supports_continuation(model) is True
+
 class _CopilotEndpointRow(BaseModel):
     model_config = ConfigDict(frozen=True, extra="ignore")
 

tests/unit/test_utils.py

@@ -476,6 +476,20 @@ def test_supports_video_input_reads_its_own_cost_map_flag(
     assert litellm.get_model_info(model)["supports_video_input"] is video_input
 
 
+
+@pytest.mark.parametrize("declared", (True, False, None))
+def test_continuation_capability_reads_declared_model_info(
+    local_model_cost_map: None, monkeypatch: pytest.MonkeyPatch, declared: bool | None
+) -> None:
+    model: Final = "openai/continuation-capability-test"
+    monkeypatch.setitem(
+        litellm.model_cost,
+        model,
+        {"litellm_provider": "openai", "mode": "chat", "supports_continuation": declared},
+    )
+    assert litellm.get_model_info(model)["supports_continuation"] is declared
+    assert litellm.supports_continuation(model) is (declared is True)
+
 def test_get_optional_params_image_gen():
     from litellm.llms.azure.image_generation import AzureGPTImageGenerationConfig
 
@@ -1175,6 +1189,7 @@ _COST_MAP_ROW_SCHEMA: Final = {
             "supports_tool_search": {"type": "boolean"},
             "supports_video_input": {"type": "boolean"},
             "supports_box_questions": {"type": "boolean"},
+            "supports_continuation": {"type": "boolean"},
             "supports_vision": {"type": "boolean"},
             "supports_web_search": {"type": "boolean"},
             "token_count_framing": {"type": "integer"},