diff --git a/tests/test_litellm/integrations/test_prometheus_latency_buckets.py b/tests/test_litellm/integrations/test_prometheus_latency_buckets.py
new file mode 100644
index 00000000000..3160bb5a9de
--- /dev/null
+++ b/tests/test_litellm/integrations/test_prometheus_latency_buckets.py
@@ -0,0 +1,144 @@
+"""
+Tests that prometheus_latency_buckets is respected end-to-end:
+- histograms are registered with the custom boundaries
+- observed values land in the expected bucket
+"""
+
+from datetime import datetime, timedelta
+
+import pytest
+from prometheus_client import REGISTRY
+
+import litellm
+from litellm.integrations.prometheus import PrometheusLogger
+from litellm.types.integrations.prometheus import UserAPIKeyLabelValues
+
+
+@pytest.fixture(autouse=True)
+def reset_prometheus_registry():
+    collectors = list(REGISTRY._collector_to_names.keys())
+    for c in collectors:
+        REGISTRY.unregister(c)
+    yield
+    collectors = list(REGISTRY._collector_to_names.keys())
+    for c in collectors:
+        REGISTRY.unregister(c)
+
+
+@pytest.fixture(autouse=True)
+def reset_litellm_buckets():
+    original = litellm.prometheus_latency_buckets
+    yield
+    litellm.prometheus_latency_buckets = original
+
+
+def _make_enum_values() -> UserAPIKeyLabelValues:
+    return UserAPIKeyLabelValues(
+        end_user=None,
+        hashed_api_key="test-key",
+        api_key_alias="test-alias",
+        requested_model="gpt-4",
+        model_group="gpt-4",
+        team=None,
+        team_alias=None,
+        user=None,
+        user_email=None,
+        status_code="200",
+        model="gpt-4",
+        litellm_model_name="gpt-4",
+        tags=[],
+        model_id="model-id-1",
+        api_base="https://api.openai.com",
+        api_provider="openai",
+        exception_status=None,
+        exception_class=None,
+        custom_metadata_labels={},
+        route=None,
+    )
+
+
+def _get_bucket_count(metric_name: str, le: str) -> float:
+    """Return the bucket count for a specific upper bound from the live registry."""
+    for metric_family in REGISTRY.collect():
+        if metric_family.name == metric_name:
+            for sample in metric_family.samples:
+                if (
+                    sample.name == f"{metric_name}_bucket"
+                    and sample.labels.get("le") == le
+                ):
+                    return sample.value
+    raise AssertionError(f"Bucket le={le} not found in metric {metric_name}")
+
+
+def test_custom_latency_buckets_registered():
+    """Histograms use prometheus_latency_buckets when set."""
+    litellm.prometheus_latency_buckets = (1.0, 5.0, float("inf"))
+
+    logger = PrometheusLogger()
+
+    assert logger._get_latency_buckets() == (1.0, 5.0, float("inf"))
+    assert logger.litellm_request_total_latency_metric._upper_bounds == [
+        1.0,
+        5.0,
+        float("inf"),
+    ]
+    assert logger.litellm_llm_api_latency_metric._upper_bounds == [
+        1.0,
+        5.0,
+        float("inf"),
+    ]
+
+
+def test_observation_lands_in_correct_custom_bucket():
+    """
+    With buckets [1.0, 5.0, inf], a 2-second observation must:
+    - NOT be counted in the le=1.0 bucket
+    - BE counted in the le=5.0 bucket
+    - BE counted in the le=+Inf bucket
+    """
+    litellm.prometheus_latency_buckets = (1.0, 5.0, float("inf"))
+
+    logger = PrometheusLogger()
+    enum_values = _make_enum_values()
+
+    now = datetime.now()
+    kwargs = {
+        "start_time": now - timedelta(seconds=2),
+        "end_time": now,
+        "api_call_start_time": now - timedelta(seconds=2),
+        "litellm_params": {"metadata": {}},
+        "model": "gpt-4",
+    }
+
+    logger._set_latency_metrics(
+        kwargs=kwargs,
+        model="gpt-4",
+        user_api_key="test-key",
+        user_api_key_alias="test-alias",
+        user_api_team=None,
+        user_api_team_alias=None,
+        enum_values=enum_values,
+    )
+
+    # le=1.0 bucket must be 0 (2s > 1.0)
+    assert _get_bucket_count("litellm_request_total_latency_metric", "1.0") == 0.0
+
+    # le=5.0 bucket must be 1 (2s <= 5.0)
+    assert _get_bucket_count("litellm_request_total_latency_metric", "5.0") == 1.0
+
+    # le=+Inf must also be 1
+    assert _get_bucket_count("litellm_request_total_latency_metric", "+Inf") == 1.0
+
+
+def test_default_buckets_used_when_not_set():
+    """When prometheus_latency_buckets is None, the module-level LATENCY_BUCKETS constant is used."""
+    from litellm.types.integrations.prometheus import LATENCY_BUCKETS
+
+    litellm.prometheus_latency_buckets = None
+
+    logger = PrometheusLogger()
+
+    assert logger._get_latency_buckets() == LATENCY_BUCKETS
+    assert logger.litellm_request_total_latency_metric._upper_bounds == list(
+        LATENCY_BUCKETS
+    )