diff --git a/.buildkite/test-amd.yaml b/.buildkite/test-amd.yaml
index 9c1b2077a7e6..61d43db4b089 100644
--- a/.buildkite/test-amd.yaml
+++ b/.buildkite/test-amd.yaml
@@ -380,6 +380,7 @@ steps:
   - tests/test_outputs.py
   - tests/test_pooling_params.py
   - tests/test_ray_env.py
+  - tests/test_sampling_params.py
   - tests/multimodal
   - tests/renderers
   - tests/standalone_tests/lazy_imports.py
@@ -395,6 +396,7 @@ steps:
   - pytest -v -s test_outputs.py
   - pytest -v -s test_pooling_params.py
   - pytest -v -s test_ray_env.py
+  - pytest -v -s test_sampling_params.py
   - pytest -v -s -m 'cpu_test' multimodal
   - pytest -v -s renderers
   - pytest -v -s tokenizers_
diff --git a/.buildkite/test_areas/misc.yaml b/.buildkite/test_areas/misc.yaml
index 7aeb84050660..fd6ef2e61bae 100644
--- a/.buildkite/test_areas/misc.yaml
+++ b/.buildkite/test_areas/misc.yaml
@@ -351,6 +351,7 @@ steps:
   - tests/test_outputs.py
   - tests/test_pooling_params.py
   - tests/test_ray_env.py
+  - tests/test_sampling_params.py
   - tests/multimodal
   - tests/renderers
   - tests/standalone_tests/lazy_imports.py
@@ -368,6 +369,7 @@ steps:
   - pytest -v -s test_outputs.py
   - pytest -v -s test_pooling_params.py
   - pytest -v -s test_ray_env.py
+  - pytest -v -s test_sampling_params.py
   - pytest -v -s -m 'cpu_test' multimodal
   - pytest -v -s renderers
   - pytest -v -s reasoning
diff --git a/tests/test_sampling_params.py b/tests/test_sampling_params.py
new file mode 100644
index 000000000000..e5d811fbb137
--- /dev/null
+++ b/tests/test_sampling_params.py
@@ -0,0 +1,50 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+from dataclasses import dataclass
+
+import pytest
+
+from vllm import SamplingParams
+
+
+@dataclass
+class MockModelConfig:
+    is_diffusion: bool = False
+    max_logprobs: int = 20
+    logits_processors: list | None = None
+
+    def get_vocab_size(self) -> int:
+        return 1024
+
+
+@pytest.mark.parametrize(
+    "kwargs",
+    [
+        {"temperature": 0.7},
+        {"temperature": 0.0},
+        {"min_p": 0.1},
+        {"seed": 42},
+        {"min_tokens": 5},
+        {"logit_bias": {0: 1.0}},
+        {"bad_words": ["foo"]},
+        {"allowed_token_ids": [0, 1]},
+    ],
+)
+def test_diffusion_rejects_unsupported_params(kwargs: dict):
+    params = SamplingParams(**kwargs)
+    with pytest.raises(ValueError, match="not yet supported with diffusion"):
+        params.verify(MockModelConfig(is_diffusion=True), None, None, None)
+
+
+def test_diffusion_accepts_default_params():
+    SamplingParams().verify(MockModelConfig(is_diffusion=True), None, None, None)
+
+
+def test_diffusion_accepts_top_k_top_p():
+    params = SamplingParams(top_p=0.9, top_k=10)
+    params.verify(MockModelConfig(is_diffusion=True), None, None, None)
+
+
+def test_non_diffusion_models_unaffected():
+    params = SamplingParams(temperature=0.7, top_k=10, seed=42)
+    params.verify(MockModelConfig(), None, None, None)
diff --git a/tests/v1/sample/test_logprobs.py b/tests/v1/sample/test_logprobs.py
index ec1502727923..fba240fea6a6 100644
--- a/tests/v1/sample/test_logprobs.py
+++ b/tests/v1/sample/test_logprobs.py
@@ -85,6 +85,7 @@ def _model_config(vocab_size: int = 10):
     return SimpleNamespace(
         max_logprobs=20,
         logits_processors=None,
+        is_diffusion=False,
         get_vocab_size=lambda: vocab_size,
     )
 
diff --git a/vllm/sampling_params.py b/vllm/sampling_params.py
index 5df2e8cfc177..f0966902d363 100644
--- a/vllm/sampling_params.py
+++ b/vllm/sampling_params.py
@@ -745,6 +745,7 @@ def verify(
         self._validate_logits_processors(model_config)
         self._validate_allowed_token_ids(tokenizer)
         self._validate_spec_decode(speculative_config)
+        self._validate_diffusion(model_config)
         self._validate_structured_outputs(
             model_config, structured_outputs_config, tokenizer
         )
@@ -878,6 +879,28 @@ def _validate_spec_decode(
                 "are not yet supported with speculative decoding."
             )
 
+    def _validate_diffusion(self, model_config: ModelConfig) -> None:
+        if not model_config.is_diffusion:
+            return
+
+        # Diffusion models denoise a whole canvas per step with a fixed
+        # temperature schedule, so per-request sampling parameters are not
+        # supported. Penalties are ignored by the sampler with a warning.
+        if (
+            self.temperature != 1.0
+            or self.min_p > _SAMPLING_EPS
+            or self.seed is not None
+            or self.min_tokens > 0
+            or self.logit_bias
+            or self.bad_words
+            or self.allowed_token_ids
+        ):
+            raise ValueError(
+                "The temperature, min_p, seed, min_tokens, logit_bias, "
+                "bad_words, and allowed_token_ids sampling parameters "
+                "are not yet supported with diffusion models."
+            )
+
     def _validate_structured_outputs(
         self,
         model_config: ModelConfig,