diff --git a/docs/user_guide/diffusion/attention_backends/huggingface_hub.md b/docs/user_guide/diffusion/attention_backends/huggingface_hub.md
index 846e69fbc..659fe99f7 100644
--- a/docs/user_guide/diffusion/attention_backends/huggingface_hub.md
+++ b/docs/user_guide/diffusion/attention_backends/huggingface_hub.md
@@ -12,7 +12,7 @@ implementations to reduce numerical drift and sampling divergence.
 ## Installation
 
 ```bash
-pip install kernels==0.14.1
+pip install kernels==0.16.0
 ```
 
 If `kernels` is unavailable, an explicit `FLASH_ATTN_HUB` or
diff --git a/requirements/common.txt b/requirements/common.txt
index 99d7412d2..908726031 100644
--- a/requirements/common.txt
+++ b/requirements/common.txt
@@ -1,7 +1,8 @@
 # Common dependencies for all platforms
 # transformers: required unified multimodal model classes are unavailable
 # before 5.10.1; 5.10.0 is yanked, and 5.15 has a known construction
-# regression. transformers 5.9 also changed
+# regression. kernels 0.16.0's stable-ABI Hub integration also requires
+# transformers 5.13.0 or newer. transformers 5.9 also changed
 # _adjust_missing_and_unexpected_keys from
 # list+list to set|set, breaking models that declare
 # _keys_to_ignore_on_load_unexpected as a list (e.g. OpenMOSS-Team/MOSS-TTS-Nano)
@@ -9,12 +10,12 @@
 # That breakage is now handled in code instead of by pinning:
 # MossTTSNanoForGeneration wraps weight loading in
 # transformers_keys_to_ignore_compat (vllm_omni/model_executor/models/utils.py).
-transformers >= 5.10.1, < 5.15
+transformers >= 5.13.0, < 5.15
 av>=14.0.0
 omegaconf>=2.3.0
 diffusers==0.40.0
 # Hub kernels, including LTX-2.5 NATTEN
-kernels==0.14.1
+kernels==0.16.0
 safetensors>=0.8.0
 accelerate==1.12.0
 soundfile>=0.13.1
diff --git a/tests/config/test_dependency_constraints.py b/tests/config/test_dependency_constraints.py
new file mode 100644
index 000000000..cb7269de5
--- /dev/null
+++ b/tests/config/test_dependency_constraints.py
@@ -0,0 +1,20 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
+from __future__ import annotations
+
+import re
+from pathlib import Path
+
+import pytest
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
+
+_COMMON_REQUIREMENTS = Path(__file__).parents[2] / "requirements" / "common.txt"
+
+
+def test_common_requirements_keep_hub_kernel_dependencies_compatible():
+    requirements = _COMMON_REQUIREMENTS.read_text()
+
+    assert re.search(r"^transformers\s*>=\s*5\.13\.0,\s*<\s*5\.15\s*$", requirements, re.MULTILINE)
+    assert re.search(r"^kernels\s*==\s*0\.16\.0\s*$", requirements, re.MULTILINE)