diff --git a/headroom/backends/litellm.py b/headroom/backends/litellm.py
index 36987b3e5..08e972908 100644
--- a/headroom/backends/litellm.py
+++ b/headroom/backends/litellm.py
@@ -445,7 +445,12 @@ def _anthropic_usage_from_litellm(litellm_usage: Any) -> dict[str, Any]:
     prompt_tokens = int(getattr(litellm_usage, "prompt_tokens", 0) or 0)
     usage: dict[str, Any] = {
         "input_tokens": max(prompt_tokens - cache_read - cache_write, 0),
-        "output_tokens": getattr(litellm_usage, "completion_tokens", 0),
+        # None-guard like the other fields: LiteLLM's Usage always carries the
+        # completion_tokens attribute, so the getattr default never fires, but a
+        # provider can leave it None. Emitting output_tokens=None would break the
+        # RequestOutcome int contract downstream (e.g. prometheus does
+        # tokens_output_total += output_tokens -> TypeError).
+        "output_tokens": int(getattr(litellm_usage, "completion_tokens", 0) or 0),
     }
     if cache_read or cache_write:
         usage["cache_read_input_tokens"] = cache_read
diff --git a/tests/test_litellm_nonstream_cache_usage.py b/tests/test_litellm_nonstream_cache_usage.py
index 872456bcf..1943669b6 100644
--- a/tests/test_litellm_nonstream_cache_usage.py
+++ b/tests/test_litellm_nonstream_cache_usage.py
@@ -72,3 +72,15 @@ def test_input_tokens_never_negative() -> None:
         )
     )
     assert usage["input_tokens"] == 0
+
+
+def test_output_tokens_none_coerced_to_zero() -> None:
+    # A provider can carry the completion_tokens attribute but leave it None.
+    # The mapping must emit an int (0), not None, so RequestOutcome's int
+    # contract holds downstream (prometheus does tokens_output_total +=
+    # output_tokens, which would raise TypeError on None).
+    usage = _anthropic_usage_from_litellm(
+        SimpleNamespace(prompt_tokens=100, completion_tokens=None)
+    )
+    assert usage["output_tokens"] == 0
+    assert isinstance(usage["output_tokens"], int)