diff --git a/resources_servers/code_gen/app.py b/resources_servers/code_gen/app.py
index d6787203db..317da60e30 100644
--- a/resources_servers/code_gen/app.py
+++ b/resources_servers/code_gen/app.py
@@ -64,6 +64,7 @@ class CompCodingRunRequest(BaseRunRequest):
 
 class CompCodingVerifyRequest(CompCodingRunRequest, BaseVerifyRequest):
     verifier_metadata: Optional[Dict[str, Any]] = None
+    target_language: Optional[str] = None
 
 
 class CompCodingVerifyResponse(BaseVerifyResponse):
@@ -74,6 +75,7 @@ class CompCodingVerifyResponse(BaseVerifyResponse):
     unit_tests_time_taken: Optional[float] = None
     reasoning_format_violation_rate: float = 0.0
     difficulty: Optional[str] = None
+    target_language: Optional[str] = None
 
 
 # ----------------------------
@@ -124,6 +126,7 @@ def compute_metrics(self, tasks: List[List[Dict[str, Any]]]) -> Dict[str, Any]:
         )
         add_avg_sample_std_dev(metrics, all_score_dicts, score_names, max_k)
         metrics.update(compute_subset_metrics(tasks, "difficulty", self._code_score_fn, "extracted_model_code"))
+        metrics.update(compute_subset_metrics(tasks, "target_language", self._code_score_fn, "extracted_model_code"))
         return metrics
 
     def get_key_metrics(self, agent_metrics: Dict[str, Any]) -> Dict[str, Any]:
diff --git a/resources_servers/code_gen/tests/test_app.py b/resources_servers/code_gen/tests/test_app.py
index e39c287eda..5c601911bc 100644
--- a/resources_servers/code_gen/tests/test_app.py
+++ b/resources_servers/code_gen/tests/test_app.py
@@ -146,6 +146,10 @@ def test_verify_missing_response_validation_error(self) -> None:
                 verifier_metadata={"unit_tests": {"inputs": ["1\n"], "outputs": ["1"]}},
             )
 
+    def test_verify_target_language_field(self) -> None:
+        resp = CompCodingVerifyResponse(reward=0.0, target_language="ja")
+        assert resp.target_language == "ja"
+
     async def test_verify_no_code_block(self, code_gen_resources_server_client: TestClient) -> None:
         """Test when response contains no code block - should extract raw text"""
         response = NeMoGymResponse(
@@ -327,6 +331,23 @@ def test_no_answer_tracked(self) -> None:
         assert "pass@1[avg-of-2]/no_answer" in m
         assert m["pass@1[avg-of-2]/no_answer"] == pytest.approx(50.0)
 
+    def test_produces_per_language_subsets(self):
+        server = _make_server()
+        tasks = [
+            [
+                {"reward": 1.0, "extracted_model_code": "print(1)", "target_language": "de"},
+                {"reward": 1.0, "extracted_model_code": "print(1)", "target_language": "de"},
+            ],
+            [
+                {"reward": 0.0, "extracted_model_code": "print(2)", "target_language": "fr"},
+                {"reward": 0.0, "extracted_model_code": "print(2)", "target_language": "fr"},
+            ],
+        ]
+        m = server.compute_metrics(tasks)
+        assert "de/pass@1/accuracy" in m
+        assert "fr/pass@1/accuracy" in m
+        assert m["de/pass@1/accuracy"] > m["fr/pass@1/accuracy"]
+
 
 class TestGetKeyMetrics:
     @pytest.mark.asyncio