From 3e654abaf1ed124c9cd03df5657e7bf9e8832abc Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Mon, 13 Jul 2026 15:41:23 +0000
Subject: [PATCH] fix(router): stop excluding a deployment one request before
 its rpm cap

---
 litellm/router_strategy/lowest_tpm_rpm.py     |  2 +-
 litellm/router_strategy/lowest_tpm_rpm_v2.py  |  2 +-
 .../local_testing/test_tpm_rpm_routing_v2.py  | 88 +++++++++++++++++++
 3 files changed, 90 insertions(+), 2 deletions(-)

diff --git a/litellm/router_strategy/lowest_tpm_rpm.py b/litellm/router_strategy/lowest_tpm_rpm.py
index 89ad7526f20..40eef179b54 100644
--- a/litellm/router_strategy/lowest_tpm_rpm.py
+++ b/litellm/router_strategy/lowest_tpm_rpm.py
@@ -224,7 +224,7 @@ class LowestTPMLoggingHandler(CustomLogger):
 
             if item_tpm + input_tokens > _deployment_tpm:
                 continue
-            elif (rpm_dict is not None and item in rpm_dict) and (rpm_dict[item] + 1 >= _deployment_rpm):
+            elif (rpm_dict is not None and item in rpm_dict) and (rpm_dict[item] + 1 > _deployment_rpm):
                 continue
             elif item_tpm < lowest_tpm:
                 lowest_tpm = item_tpm
diff --git a/litellm/router_strategy/lowest_tpm_rpm_v2.py b/litellm/router_strategy/lowest_tpm_rpm_v2.py
index d63621a7d77..f7dfaf5c1ea 100644
--- a/litellm/router_strategy/lowest_tpm_rpm_v2.py
+++ b/litellm/router_strategy/lowest_tpm_rpm_v2.py
@@ -357,7 +357,7 @@ class LowestTPMLoggingHandler_v2(BaseRoutingStrategy, CustomLogger):
             elif (
                 (rpm_dict is not None and item in rpm_dict)
                 and rpm_dict[item] is not None
-                and (rpm_dict[item] + 1 >= _deployment_rpm)
+                and (rpm_dict[item] + 1 > _deployment_rpm)
             ):
                 continue
             elif item_tpm == lowest_tpm:
diff --git a/tests/local_testing/test_tpm_rpm_routing_v2.py b/tests/local_testing/test_tpm_rpm_routing_v2.py
index 211af566424..0db6df28009 100644
--- a/tests/local_testing/test_tpm_rpm_routing_v2.py
+++ b/tests/local_testing/test_tpm_rpm_routing_v2.py
@@ -725,6 +725,94 @@ def test_return_potential_deployments():
     assert len(potential_deployments) == 1
 
 
+def test_return_potential_deployments_one_request_before_rpm_limit():
+    """
+    Regression test for https://github.com/BerriAI/litellm/issues/33032
+
+    A deployment whose current rpm usage is one below its rpm cap must NOT be
+    filtered out. Routing there makes usage reach (not exceed) the limit
+    """
+
+    test_cache = DualCache()
+    lowest_tpm_logger = LowestTPMLoggingHandler(router_cache=test_cache)
+
+    deployment_id = "dd8e67fce56963bae6a60206b48d3f03faeb43be20cf0fd96a5f39b1a2bbd11d"
+    args: Dict = {
+        "healthy_deployments": [
+            {
+                "model_name": "model-test",
+                "litellm_params": {
+                    "rpm": 10,
+                    "api_key": "sk-1234",
+                    "model": "openai/gpt-3.5-turbo",
+                    "mock_response": "Hello, world!",
+                },
+                "model_info": {
+                    "id": deployment_id,
+                    "db_model": False,
+                },
+            },
+        ],
+        "all_deployments": {
+            deployment_id: None,
+            f"{deployment_id}:tpm:02-17": 0,
+        },
+        "input_tokens": 98,
+        "rpm_dict": {
+            deployment_id: 9,
+        },
+    }
+
+    potential_deployments = lowest_tpm_logger._return_potential_deployments(
+        healthy_deployments=args["healthy_deployments"],
+        all_deployments=args["all_deployments"],
+        input_tokens=args["input_tokens"],
+        rpm_dict=args["rpm_dict"],
+    )
+
+    assert len(potential_deployments) == 1
+
+
+def test_get_available_deployment_one_request_before_rpm_limit():
+    """
+    Regression test for https://github.com/BerriAI/litellm/issues/33032 on the
+    sync usage-based-routing (v1) handler
+    """
+    from litellm.router_strategy.lowest_tpm_rpm import (
+        LowestTPMLoggingHandler as LowestTPMLoggingHandlerV1,
+    )
+
+    model_group = "model-test"
+    deployment_id = "abc-123"
+    test_cache = DualCache()
+    handler = LowestTPMLoggingHandlerV1(router_cache=test_cache)
+
+    current_minute = datetime.now().strftime("%H-%M")
+    test_cache.set_cache(
+        key=f"{model_group}:tpm:{current_minute}", value={deployment_id: 0}
+    )
+    test_cache.set_cache(
+        key=f"{model_group}:rpm:{current_minute}", value={deployment_id: 9}
+    )
+
+    healthy_deployments = [
+        {
+            "model_name": model_group,
+            "litellm_params": {"rpm": 10, "model": "openai/gpt-3.5-turbo"},
+            "model_info": {"id": deployment_id},
+        }
+    ]
+
+    deployment = handler.get_available_deployments(
+        model_group=model_group,
+        healthy_deployments=healthy_deployments,
+        messages=[{"role": "user", "content": "hi"}],
+    )
+
+    assert deployment is not None
+    assert deployment["model_info"]["id"] == deployment_id
+
+
 @pytest.mark.asyncio
 async def test_tpm_rpm_routing_model_name_checks():
     deployment = {