fix: align request usage entries with known token usage

src/agents/run_internal/model_retry.py

@@ -315,16 +315,6 @@ async def _await_model_attempt(
     raise timeout_error from None
 
 
-def _build_zero_request_usage_entry() -> RequestUsage:
-    return RequestUsage(
-        input_tokens=0,
-        output_tokens=0,
-        total_tokens=0,
-        input_tokens_details=Usage().input_tokens_details,
-        output_tokens_details=Usage().output_tokens_details,
-    )
-
-
 def _build_request_usage_entry_from_usage(usage: Usage) -> RequestUsage:
     return RequestUsage(
         input_tokens=usage.input_tokens,
@@ -339,14 +329,14 @@ def apply_retry_attempt_usage(usage: Usage, failed_attempts: int) -> Usage:
     if failed_attempts <= 0:
         return usage
 
-    successful_request_entries = list(usage.request_usage_entries)
-    if not successful_request_entries:
-        successful_request_entries.append(_build_request_usage_entry_from_usage(usage))
+    request_entries = list(usage.request_usage_entries)
+    if not request_entries and usage.total_tokens > 0:
+        request_entries.append(_build_request_usage_entry_from_usage(usage))
 
     usage.requests = max(usage.requests, 1) + failed_attempts
-    usage.request_usage_entries = [
-        _build_zero_request_usage_entry() for _ in range(failed_attempts)
-    ] + successful_request_entries
+    # Failed attempts do not carry provider usage. Keep them in the aggregate request count,
+    # but do not synthesize zero-valued entries that look like known per-request usage.
+    usage.request_usage_entries = request_entries
     return usage
 
 

src/agents/usage.py

@@ -218,8 +218,9 @@ class Usage:
     request_usage_entries: list[RequestUsage] = field(default_factory=list)
     """List of RequestUsage entries for accurate per-request cost calculation.
 
-    Each call to `add()` automatically creates an entry in this list if the added usage
-    represents a new request (i.e., has non-zero tokens).
+    Entries are created only when token usage is known for an individual request. A request whose
+    provider reports no usage still increments `requests`, but it does not create a synthetic
+    zero-valued entry. As a result, this list can contain fewer entries than `requests`.
 
     Example:
         For a run that makes 3 API calls with 100K, 150K, and 80K input tokens each,

tests/models/test_model_retry.py

@@ -921,15 +921,14 @@ async def get_response() -> ModelResponse:
     )
 
     assert result.usage.requests == 2
-    assert len(result.usage.request_usage_entries) == 2
-    assert result.usage.request_usage_entries[0].total_tokens == 0
-    assert result.usage.request_usage_entries[1].input_tokens == 11
-    assert result.usage.request_usage_entries[1].output_tokens == 7
-    assert result.usage.request_usage_entries[1].total_tokens == 18
+    assert len(result.usage.request_usage_entries) == 1
+    assert result.usage.request_usage_entries[0].input_tokens == 11
+    assert result.usage.request_usage_entries[0].output_tokens == 7
+    assert result.usage.request_usage_entries[0].total_tokens == 18
 
 
 @pytest.mark.asyncio
-async def test_get_response_with_retry_preserves_zero_token_successful_request_usage_entry(
+async def test_get_response_with_retry_omits_unknown_successful_request_usage_entry(
     monkeypatch,
 ) -> None:
     calls = 0
@@ -967,9 +966,7 @@ async def get_response() -> ModelResponse:
     )
 
     assert result.usage.requests == 2
-    assert len(result.usage.request_usage_entries) == 2
-    assert result.usage.request_usage_entries[0].total_tokens == 0
-    assert result.usage.request_usage_entries[1].total_tokens == 0
+    assert result.usage.request_usage_entries == []
 
 
 @pytest.mark.asyncio
@@ -1126,9 +1123,8 @@ async def get_response() -> ModelResponse:
     assert rewinds == 1
     assert sleeps == [1.0]
     assert result.usage.requests == 2
-    assert len(result.usage.request_usage_entries) == 2
-    assert result.usage.request_usage_entries[0].total_tokens == 0
-    assert result.usage.request_usage_entries[1].total_tokens == 5
+    assert len(result.usage.request_usage_entries) == 1
+    assert result.usage.request_usage_entries[0].total_tokens == 5
 
 
 @pytest.mark.asyncio

tests/test_agent_runner_streamed.py

@@ -431,11 +431,10 @@ async def test_streamed_run_preserves_request_usage_entries_after_retry() -> Non
 
     usage = result.context_wrapper.usage
     assert usage.requests == 2
-    assert len(usage.request_usage_entries) == 2
-    assert usage.request_usage_entries[0].total_tokens == 0
-    assert usage.request_usage_entries[1].input_tokens == 10
-    assert usage.request_usage_entries[1].output_tokens == 5
-    assert usage.request_usage_entries[1].total_tokens == 15
+    assert len(usage.request_usage_entries) == 1
+    assert usage.request_usage_entries[0].input_tokens == 10
+    assert usage.request_usage_entries[0].output_tokens == 5
+    assert usage.request_usage_entries[0].total_tokens == 15
 
 
 @pytest.mark.asyncio
@@ -483,9 +482,7 @@ async def missing_usage_stream(_call) -> AsyncIterator[TResponseStreamEvent]:
     usage = result.context_wrapper.usage
     assert len(model.calls) == 2
     assert usage.requests == 2
-    assert len(usage.request_usage_entries) == 2
-    assert usage.request_usage_entries[0].total_tokens == 0
-    assert usage.request_usage_entries[1].total_tokens == 0
+    assert usage.request_usage_entries == []
 
 
 @pytest.mark.asyncio
@@ -557,11 +554,10 @@ async def test_streamed_run_preserves_request_usage_entries_after_conversation_l
 
     usage = result.context_wrapper.usage
     assert usage.requests == 2
-    assert len(usage.request_usage_entries) == 2
-    assert usage.request_usage_entries[0].total_tokens == 0
-    assert usage.request_usage_entries[1].input_tokens == 10
-    assert usage.request_usage_entries[1].output_tokens == 5
-    assert usage.request_usage_entries[1].total_tokens == 15
+    assert len(usage.request_usage_entries) == 1
+    assert usage.request_usage_entries[0].input_tokens == 10
+    assert usage.request_usage_entries[0].output_tokens == 5
+    assert usage.request_usage_entries[0].total_tokens == 15
 
 
 @pytest.mark.allow_call_model_methods

tests/test_usage.py

@@ -223,7 +223,8 @@ def test_usage_add_ignores_zero_token_requests():
 
     u1.add(u2)
 
-    # Should not create a request_usage_entry for zero tokens
+    assert u1.requests == 1
+    # The aggregate request count remains authoritative when token usage is unavailable.
     assert len(u1.request_usage_entries) == 0