diff --git a/RELEASE.md b/RELEASE.md
index 9f13364..f9d2e2a 100644
--- a/RELEASE.md
+++ b/RELEASE.md
@@ -4,6 +4,12 @@ OpenFilter Library release notes
 
 ## [Unreleased]
 
+## v1.1.2 - 2026-06-28
+
+### Fixed
+
+- **GPU usage percent emitted as float for Cloud Monitoring ([FILTER-585](https://plainsight-ai.atlassian.net/browse/FILTER-585))**: `Metrics.gpu_thread_func` now casts `gpu_usage_percent` to `float` before export. NVML reports `gpu_util` as an integer, but the Cloud Monitoring descriptor `openfilter_gpu_usage_percent` is type-locked to DOUBLE, so each export batch had one point rejected with `value type for metric must be DOUBLE, but is INT64`. The cast is targeted to this single summary field; the per-device `gpuN` series and the legitimately INT64 descriptors (`gpu_accessible`, `camera_connected`) are unchanged.
+
 ## v1.1.1 - 2026-06-01
 
 ### Fixed
diff --git a/VERSION b/VERSION
index 56130fb..0f1acbd 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-v1.1.1
+v1.1.2
diff --git a/openfilter/filter_runtime/metrics.py b/openfilter/filter_runtime/metrics.py
index edbc201..2990157 100644
--- a/openfilter/filter_runtime/metrics.py
+++ b/openfilter/filter_runtime/metrics.py
@@ -134,7 +134,7 @@ def gpu_thread_func(self, stop_evt: Event):
 
                 self.gpu = gpu
                 self.gpu_accessible = 1
-                self.gpu_usage_percent = gpu.get('gpu0', 0.0)
+                self.gpu_usage_percent = float(gpu.get('gpu0', 0.0))  # NVML gpu_util is int; Cloud Monitoring's DOUBLE descriptor rejects it otherwise
 
         except Exception as exc:
             logger.error(exc)
diff --git a/tests/test_gpu_usage_percent_float.py b/tests/test_gpu_usage_percent_float.py
new file mode 100644
index 0000000..4f9d524
--- /dev/null
+++ b/tests/test_gpu_usage_percent_float.py
@@ -0,0 +1,35 @@
+"""Tests for ensuring that openfilter_gpu_usage_percent is always emitted as a float."""
+
+from unittest.mock import MagicMock, patch
+
+from openfilter.filter_runtime.gpu import GPUMetrics
+from openfilter.filter_runtime.metrics import Metrics
+
+
+def _run_one_iteration(mock_metrics):
+    """Run gpu_thread_func for exactly one loop iteration with mocked NVML metrics."""
+    metrics = object.__new__(Metrics)  # bare object, skip __init__'s background threads
+
+    stop_evt = MagicMock()
+    stop_evt.wait.side_effect = [False, True]  # run the loop body exactly once
+
+    with patch('openfilter.filter_runtime.metrics.get_gpu_metrics', return_value=mock_metrics):
+        metrics.gpu_thread_func(stop_evt)
+
+    return metrics
+
+
+def test_gpu_usage_percent_is_float():
+    """gpu_usage_percent is cast to float even when NVML returns an int gpu_util."""
+    metrics = _run_one_iteration([GPUMetrics(index=0, gpu_util=55, mem_used_mb=1024)])
+
+    assert type(metrics.gpu_usage_percent) is float
+    assert metrics.gpu_usage_percent == 55.0
+
+
+def test_gpu_usage_percent_is_float_when_gpu_idle():
+    """An idle GPU reports integer gpu_util==0; without the float() cast this path would emit an int 0 and be rejected by the DOUBLE-locked Cloud Monitoring descriptor, so this test fails on the un-cast code."""
+    metrics = _run_one_iteration([GPUMetrics(index=0, gpu_util=0, mem_used_mb=1024)])
+
+    assert type(metrics.gpu_usage_percent) is float
+    assert metrics.gpu_usage_percent == 0.0