diff --git a/openevolve/config.py b/openevolve/config.py
index 4fdab7998..2014802ff 100644
--- a/openevolve/config.py
+++ b/openevolve/config.py
@@ -145,7 +145,10 @@ def __post_init__(self):
         if self.primary_model:
             # Create primary model
             primary_model = LLMModelConfig(
-                name=self.primary_model, weight=self.primary_model_weight or 1.0
+                name=self.primary_model,
+                weight=(
+                    self.primary_model_weight if self.primary_model_weight is not None else 1.0
+                ),
             )
             self.models.append(primary_model)
 
@@ -218,7 +221,10 @@ def rebuild_models(self) -> None:
         if self.primary_model:
             # Create primary model
             primary_model = LLMModelConfig(
-                name=self.primary_model, weight=self.primary_model_weight or 1.0
+                name=self.primary_model,
+                weight=(
+                    self.primary_model_weight if self.primary_model_weight is not None else 1.0
+                ),
             )
             self.models.append(primary_model)
 
diff --git a/tests/test_primary_model_weights.py b/tests/test_primary_model_weights.py
new file mode 100644
index 000000000..291b85a96
--- /dev/null
+++ b/tests/test_primary_model_weights.py
@@ -0,0 +1,95 @@
+"""Legacy model weights must retain the ensemble's weighted-selection semantics."""
+
+import tempfile
+import unittest
+from pathlib import Path
+
+import yaml
+
+from openevolve.config import Config, LLMConfig
+from openevolve.llm.base import LLMInterface
+from openevolve.llm.ensemble import LLMEnsemble
+
+
+class RecordingLLM(LLMInterface):
+    def __init__(self, model, calls):
+        self.model = model
+        self.calls = calls
+
+    async def generate(self, prompt, **kwargs):
+        self.calls.append(self.model)
+        return self.model
+
+    async def generate_with_context(self, system_message, messages, **kwargs):
+        self.calls.append(self.model)
+        return self.model
+
+
+class TestPrimaryModelWeights(unittest.IsolatedAsyncioTestCase):
+    async def assert_weighted_generation(self, config, expected_weight):
+        self.assertEqual([model.name for model in config.models], ["primary", "secondary"])
+        calls = []
+
+        def create_client(model_config):
+            return RecordingLLM(model_config.name, calls)
+
+        for model in config.models:
+            model.init_client = create_client
+
+        ensemble = LLMEnsemble(config.models)
+        # Seed 1 draws 0.134... first. With a positive primary weight of 0.25
+        # or 1.0 it selects primary; a zero weight must always select secondary.
+        expected_model = "secondary" if expected_weight == 0 else "primary"
+        result = await ensemble.generate_with_context(
+            "system", [{"role": "user", "content": "prompt"}]
+        )
+        self.assertEqual(result, expected_model)
+        self.assertEqual(calls, [expected_model])
+        self.assertEqual([model.weight for model in config.models], [expected_weight, 1.0])
+        self.assertEqual(
+            [model.weight for model in config.evaluator_models], [expected_weight, 1.0]
+        )
+
+    async def test_constructor_preserves_primary_weight(self):
+        for weight in (None, 0.0, 0.25):
+            with self.subTest(weight=weight):
+                config = LLMConfig(
+                    primary_model="primary",
+                    primary_model_weight=weight,
+                    secondary_model="secondary",
+                    secondary_model_weight=1.0,
+                    random_seed=1,
+                )
+                await self.assert_weighted_generation(config, 1.0 if weight is None else weight)
+
+    async def test_yaml_preserves_primary_weight(self):
+        for weight in (None, 0.0, 0.25):
+            with self.subTest(weight=weight):
+                llm = {
+                    "primary_model": "primary",
+                    "secondary_model": "secondary",
+                    "secondary_model_weight": 1.0,
+                    "random_seed": 1,
+                }
+                if weight is not None:
+                    llm["primary_model_weight"] = weight
+                with tempfile.TemporaryDirectory() as directory:
+                    path = Path(directory) / "config.yaml"
+                    path.write_text(yaml.safe_dump({"llm": llm}), encoding="utf-8")
+                    config = Config.from_yaml(path).llm
+                await self.assert_weighted_generation(config, 1.0 if weight is None else weight)
+
+    async def test_rebuild_preserves_primary_weight(self):
+        for weight in (None, 0.0, 0.25):
+            with self.subTest(weight=weight):
+                config = LLMConfig(random_seed=1)
+                config.primary_model = "primary"
+                config.primary_model_weight = weight
+                config.secondary_model = "secondary"
+                config.secondary_model_weight = 1.0
+                config.rebuild_models()
+                await self.assert_weighted_generation(config, 1.0 if weight is None else weight)
+
+
+if __name__ == "__main__":
+    unittest.main()