diff --git a/litellm/router_rust/invoke/streaming.py b/litellm/router_rust/invoke/streaming.py
index d8b4f87501a..9286e9aa06c 100644
--- a/litellm/router_rust/invoke/streaming.py
+++ b/litellm/router_rust/invoke/streaming.py
@@ -684,7 +684,10 @@ def _mid_stream_continuation_eligible(
     lookup_groups: Final = fallback_lookup_groups(
         request, requested_group if isinstance(requested_group, str) else None
     )
-    targets: Final = continuation_fallback_targets(core.settings().fallbacks, lookup_groups)  # pyright: ignore[reportUnknownMemberType, reportUnknownArgumentType]  # fallbacks as Python spells them
+    targets: Final = continuation_fallback_targets(
+        request.get("fallbacks", core.settings().fallbacks),  # pyright: ignore[reportUnknownMemberType, reportUnknownArgumentType]  # the request's list replaces the router's
+        lookup_groups,
+    )
     if any(fallback_target_overrides_request_params(target) for target in targets):
         return False
     if any(_names_deployment_that_cannot_continue(core, target) for target in targets):
diff --git a/tests/unit/test_router/core/test_streaming_characterization_chat.py b/tests/unit/test_router/core/test_streaming_characterization_chat.py
index 260ad8534cf..590fafede8e 100644
--- a/tests/unit/test_router/core/test_streaming_characterization_chat.py
+++ b/tests/unit/test_router/core/test_streaming_characterization_chat.py
@@ -516,6 +516,18 @@ async def test_continuation_declines_a_fallback_target_that_overrides_request_pa
     assert host.deployments_called() == ["a"]
 
 
+async def test_continuation_reads_the_request_fallbacks_over_the_routers(
+    make_core: MakeCore, host: ScriptedHost
+) -> None:
+    host.answer("openai/a", chat_stream("he", "llo", error=rate_limit(message="request override")))
+    host.answer(PREFILL_MODEL, chat_stream(" world"))
+    core = continuation_core(make_core, fallbacks=[{"a": ["b"]}])
+    stream = await open_stream(core, fallbacks=[{"model": "b", "messages": [{"role": "user", "content": "again"}]}])
+    with pytest.raises(litellm.RateLimitError, match="request override"):
+        await collect_texts(stream)
+    assert host.deployments_called() == ["a"]
+
+
 async def test_continuation_runs_a_plain_fallback_for_an_error_before_content(
     make_core: MakeCore, host: ScriptedHost
 ) -> None: