diff --git a/tests/models/test_transformers_parametrized.py b/tests/models/test_transformers_parametrized.py
new file mode 100644
index 000000000..f00ee9b60
--- /dev/null
+++ b/tests/models/test_transformers_parametrized.py
@@ -0,0 +1,41 @@
+import pytest
+import outlines
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from pathlib import Path
+
+
+@pytest.fixture(scope="session")
+def hf_cache_dir():
+    """Shared Hugging Face cache directory for all model downloads."""
+    path = Path(".hf_cache")
+    path.mkdir(exist_ok=True)
+    return str(path)
+
+
+@pytest.mark.parametrize(
+    "backend, model_name",
+    [
+        ("transformers", "sshleifer/tiny-gpt2"),
+        ("transformers", "distilgpt2"),
+        pytest.param("vllm", "facebook/opt-125m", marks=pytest.mark.skipif(True, reason="vLLM not available on Windows")),
+    ],
+)
+@pytest.mark.slow
+@pytest.mark.flaky(reruns=2)
+def test_parametrized_steerable_model(backend, model_name, hf_cache_dir):
+    hf_model = AutoModelForCausalLM.from_pretrained(model_name, cache_dir=hf_cache_dir)
+    tokenizer = Au
+    toTokenizer.from_pretrained(model_name, cache_dir=hf_cache_dir)
+
+    model = outlines.models.transformers.Transformers(hf_model, tokenizer)
+
+    prompt = "Hello world"
+    outputs = set()
+
+    for _ in range(3):
+        output = model.generate(prompt, max_new_tokens=10)
+        assert isinstance(output, str)
+        assert len(output.strip()) > 0
+        outputs.add(output)
+
+    assert len(outputs) >= 1