fix(provider): exclude cached tokens from Gemini input usage accounting

astrbot/core/provider/sources/gemini_source.py

@@ -451,10 +451,17 @@ def _extract_reasoning_content(self, candidate: types.Candidate) -> str:
     def _extract_usage(
         self, usage_metadata: types.GenerateContentResponseUsageMetadata
     ) -> TokenUsage:
-        """Extract usage from candidate"""
+        """Extract usage from response metadata.
+
+        `prompt_token_count` includes tokens served from cache, so subtract
+        `cached_content_token_count` to avoid double-counting cached input
+        (matching the OpenAI provider's TokenUsage accounting).
+        """
+        prompt_tokens = usage_metadata.prompt_token_count or 0
+        cached = usage_metadata.cached_content_token_count or 0
         return TokenUsage(
-            input_other=usage_metadata.prompt_token_count or 0,
-            input_cached=usage_metadata.cached_content_token_count or 0,
+            input_other=prompt_tokens - cached,
+            input_cached=cached,
             output=usage_metadata.candidates_token_count or 0,
         )
 

tests/test_gemini_source.py

@@ -131,6 +131,42 @@ def test_gemini_reasoning_only_output_is_allowed():
     )
 
 
+def test_gemini_extract_usage_excludes_cached_tokens_from_input_other():
+    provider = ProviderGoogleGenAI.__new__(ProviderGoogleGenAI)
+
+    usage_metadata = SimpleNamespace(
+        prompt_token_count=100,
+        cached_content_token_count=30,
+        candidates_token_count=50,
+    )
+
+    usage = provider._extract_usage(usage_metadata)
+
+    # prompt_token_count already includes cached tokens; input_other must
+    # exclude them so input (input_other + input_cached) is not inflated.
+    assert usage.input_other == 70
+    assert usage.input_cached == 30
+    assert usage.input == 100
+    assert usage.output == 50
+
+
+def test_gemini_extract_usage_without_cache_keeps_full_prompt_tokens():
+    provider = ProviderGoogleGenAI.__new__(ProviderGoogleGenAI)
+
+    usage_metadata = SimpleNamespace(
+        prompt_token_count=100,
+        cached_content_token_count=0,
+        candidates_token_count=20,
+    )
+
+    usage = provider._extract_usage(usage_metadata)
+
+    assert usage.input_other == 100
+    assert usage.input_cached == 0
+    assert usage.input == 100
+    assert usage.output == 20
+
+
 @pytest.mark.asyncio
 async def test_gemini_get_models_retries_transient_request_error(monkeypatch):
     monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MIN_S", 0)