diff --git a/litellm/llms/deepgram/common_utils.py b/litellm/llms/deepgram/common_utils.py
index f1759f94775..947df37bbe7 100644
--- a/litellm/llms/deepgram/common_utils.py
+++ b/litellm/llms/deepgram/common_utils.py
@@ -10,6 +10,7 @@ from litellm.constants import DEEPGRAM_DEFAULT_API_BASE, DEEPGRAM_LISTEN_DEFAULT
 from litellm.llms.base_llm.chat.transformation import BaseLLMException
 
 _WEBSOCKET_SCHEMES: Final = MappingProxyType({"https": "wss", "http": "ws", "wss": "wss", "ws": "ws"})
+DEEPGRAM_LISTEN_CALLBACK_PARAMS: Final = frozenset({"callback", "callback_method"})
 
 
 class DeepgramException(BaseLLMException):
@@ -26,6 +27,10 @@ def deepgram_listen_websocket_target(api_base: str | None, query_string: str) ->
     return f"{websocket_url}?{query}"
 
 
+def deepgram_listen_callback_params(query_string: str) -> tuple[str, ...]:
+    return tuple(sorted(DEEPGRAM_LISTEN_CALLBACK_PARAMS.intersection(httpx.QueryParams(query_string).keys())))
+
+
 def deepgram_listen_model(upstream_url: str) -> str:
     models: Final = parse_qs(urlparse(upstream_url).query).get("model")
     return models[0] if models else DEEPGRAM_LISTEN_DEFAULT_MODEL
diff --git a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py
index 1abbf90cb7a..7f9e0169fb2 100644
--- a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py
+++ b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py
@@ -36,7 +36,10 @@ from litellm.litellm_core_utils.aws_partition import get_aws_dns_suffix
 from litellm.llms.anthropic.common_utils import AnthropicModelInfo
 from litellm.llms.azure.passthrough.transformation import foreign_azure_deployment
 from litellm.llms.custom_httpx.http_handler import get_async_httpx_client
-from litellm.llms.deepgram.common_utils import deepgram_listen_websocket_target
+from litellm.llms.deepgram.common_utils import (
+    deepgram_listen_callback_params,
+    deepgram_listen_websocket_target,
+)
 from litellm.llms.nvidia_nim.passthrough.transformation import nvidia_nim_model_group_in_path
 from litellm.llms.vertex_ai.vertex_llm_base import VertexBase
 from litellm.passthrough.main import AsyncPassthroughStreamingResponse
@@ -2694,6 +2697,7 @@ async def openai_websocket_proxy_route(
 _DEEPGRAM_WS_MISSING_KEY_REASON: Final = (
     "Required 'DEEPGRAM_API_KEY' in environment to make pass-through calls to Deepgram."
 )
+_DEEPGRAM_WS_CALLBACK_REASON: Final = "Deepgram callback delivery is not supported through the proxy: remove {params}"
 
 
 @router.websocket("/deepgram/v1/listen")
@@ -2712,6 +2716,14 @@ async def deepgram_listen_websocket_route(
         return
 
     await websocket.accept(subprotocol=_negotiated_websocket_subprotocol(websocket))
+    callback_params: Final = deepgram_listen_callback_params(websocket.url.query)
+    if callback_params:
+        await websocket.close(
+            code=1008,
+            reason=_DEEPGRAM_WS_CALLBACK_REASON.format(params=", ".join(callback_params)),
+        )
+        return
+
     await relay(
         websocket=websocket,
         target=deepgram_listen_websocket_target(
diff --git a/tests/test_litellm/llms/deepgram/test_deepgram_common_utils.py b/tests/test_litellm/llms/deepgram/test_deepgram_common_utils.py
index a86cb83d628..65fbf7c7870 100644
--- a/tests/test_litellm/llms/deepgram/test_deepgram_common_utils.py
+++ b/tests/test_litellm/llms/deepgram/test_deepgram_common_utils.py
@@ -7,6 +7,7 @@ import pytest
 import litellm
 from litellm.llms.deepgram.common_utils import (
     deepgram_listen_audio_seconds,
+    deepgram_listen_callback_params,
     deepgram_listen_model,
     deepgram_listen_transcript,
     deepgram_listen_websocket_target,
@@ -68,6 +69,24 @@ def test_deepgram_listen_websocket_target(api_base: str | None, query_string: st
     assert deepgram_listen_websocket_target(api_base=api_base, query_string=query_string) == expected
 
 
+@pytest.mark.parametrize(
+    ("query_string", "expected"),
+    [
+        pytest.param("model=nova-3&encoding=linear16", (), id="no callback"),
+        pytest.param("model=nova-3&callback=https%3A%2F%2Fevil.example%2Fsink", ("callback",), id="callback"),
+        pytest.param(
+            "callback_method=put&model=nova-3&callback=wss%3A%2F%2Fevil.example",
+            ("callback", "callback_method"),
+            id="callback and method",
+        ),
+        pytest.param("model=nova-3&callback_method=put", ("callback_method",), id="method alone"),
+        pytest.param("model=nova-3&callbacks=x&my_callback=y", (), id="only exact names match"),
+    ],
+)
+def test_deepgram_listen_callback_params(query_string: str, expected: tuple[str, ...]):
+    assert deepgram_listen_callback_params(query_string) == expected
+
+
 @pytest.mark.parametrize(
     ("frames", "expected_seconds"),
     [
diff --git a/tests/test_litellm/proxy/pass_through_endpoints/test_deepgram_ws_passthrough_routes.py b/tests/test_litellm/proxy/pass_through_endpoints/test_deepgram_ws_passthrough_routes.py
index 64804e621ba..5ea2b0b8ab9 100644
--- a/tests/test_litellm/proxy/pass_through_endpoints/test_deepgram_ws_passthrough_routes.py
+++ b/tests/test_litellm/proxy/pass_through_endpoints/test_deepgram_ws_passthrough_routes.py
@@ -207,6 +207,30 @@ async def test_deepgram_listen_closes_cleanly_when_provider_credentials_missing(
     assert relay.calls == []
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "query",
+    [
+        pytest.param("model=nova-3&callback=https%3A%2F%2Fsink.example%2Fdg", id="http callback"),
+        pytest.param("callback=wss%3A%2F%2Fsink.example&callback_method=put&model=nova-3", id="ws callback"),
+    ],
+)
+async def test_deepgram_listen_rejects_callback_delivery_that_would_go_unbilled(query, monkeypatch):
+    """With ``callback`` set, Deepgram sends every Results and Metadata frame to the caller's URL and only a
+    request id down this socket, so the proxy would meter zero seconds of audio; refuse before contacting Deepgram."""
+    monkeypatch.delenv("DEEPGRAM_API_BASE", raising=False)
+    websocket = _FakeWebSocket("/deepgram/v1/listen", query)
+
+    with patch(GET_CREDENTIALS, return_value="dg-provider-key"):
+        relay = await _serve(websocket)
+
+    assert relay.calls == []
+    assert websocket.closed is not None
+    assert websocket.closed[0] == 1008
+    assert "callback" in websocket.closed[1]
+    assert "dg-provider-key" not in websocket.closed[1]
+
+
 def _app_with_relay(relay: _FakeRelay) -> FastAPI:
     app = FastAPI()
     app.include_router(router)
@@ -228,6 +252,27 @@ def test_deepgram_listen_rejects_connections_without_a_litellm_key():
     get_credentials.assert_not_called()
 
 
+def test_deepgram_listen_callback_rejection_reaches_the_client_as_a_policy_close(monkeypatch):
+    monkeypatch.delenv("DEEPGRAM_API_BASE", raising=False)
+    relay = _FakeRelay()
+    client = TestClient(_app_with_relay(relay))
+
+    with (
+        patch(GET_CREDENTIALS, return_value="dg-provider-key"),
+        patch(USER_API_KEY_AUTH, new=AsyncMock(return_value=UserAPIKeyAuth(api_key="hashed"))),
+    ):
+        with pytest.raises(WebSocketDisconnect) as disconnect:
+            with client.websocket_connect(
+                "/deepgram/v1/listen?model=nova-3&callback=https%3A%2F%2Fsink.example%2Fdg",
+                headers={"Authorization": "Bearer sk-litellm-virtual"},
+            ) as connection:
+                connection.receive_text()
+
+    assert disconnect.value.code == 1008
+    assert "callback" in disconnect.value.reason
+    assert relay.calls == []
+
+
 def test_deepgram_listen_authenticates_the_litellm_key_and_relays_to_deepgram(monkeypatch):
     monkeypatch.delenv("DEEPGRAM_API_BASE", raising=False)
     relay = _FakeRelay()