diff --git a/.agents/skills/run-e2e/SKILL.md b/.agents/skills/run-e2e/SKILL.md
index 0e4a162c..15c0b9e4 100644
--- a/.agents/skills/run-e2e/SKILL.md
+++ b/.agents/skills/run-e2e/SKILL.md
@@ -1,14 +1,18 @@
 ---
 name: run-e2e
-description: Use when the user asks to run, validate, or diagnose the AFD plugin's DeepSeek-V2-Lite end-to-end tests on GPU or Ascend NPU hardware, including PR-gate E2E, GSM8K-7 accuracy, graph, eager, DBO, or 2A2F scenarios.
+description: Use when the user asks to run, validate, or diagnose the AFD plugin's DeepSeek-V2-Lite GPU/NPU or Qwen3 MoE GPU end-to-end tests, including PR-gate E2E, GSM8K-7 accuracy, graph, eager, DBO, or 2A2F scenarios.
 ---
 
 # Run AFD E2E Tests
 
 ## Scope
 
-Run the four tests in
-tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py:
+Run one of the model suites:
+
+- `tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py` on GPU or NPU
+- `tests/e2e/models/qwen3_moe/test_qwen3_moe.py` on GPU
+
+Each suite contains four default scenarios:
 
 - baseline-graph
 - afd-eager
@@ -31,8 +35,8 @@ If both are available, ask which to use. If neither is available, stop.
 Before starting pytest, confirm:
 
 - AFD_E2E_DEVICES contains the device IDs required by test cases.
-- AFD_GPU_E2E_MODEL / AFD_NPU_E2E_MODEL is set to a local path, or the
-  environment can download `deepseek-ai/DeepSeek-V2-Lite` via huggingface_hub.
+- The backend model variable is set to a local path, or the environment can
+  download the selected suite's checkpoint via huggingface_hub.
 - The selected vllm command runs.
 - pytest, afd_plugin, lm_eval, datasets, and huggingface_hub are importable.
 - HF_HOME points to the Hugging Face cache used for GSM8K and model weights.
@@ -56,7 +60,7 @@ export HF_HOME=/path/to/huggingface
 export AFD_E2E_BACKEND=gpu
 export AFD_E2E_DEVICES=0,1,2
 # Optional if the model is already local:
-# export AFD_GPU_E2E_MODEL=/path/to/DeepSeek-V2-Lite
+# export AFD_GPU_E2E_MODEL=/path/to/model
 # Optional: export AFD_GPU_E2E_VLLM_BIN=/path/to/vllm
 ~~~
 
@@ -86,9 +90,17 @@ python -m pytest -q -s \
   "tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py::test_deepseek_v2_lite[baseline-graph]"
 ~~~
 
+For Qwen3 MoE on GPU, run:
+
+~~~bash
+python -m pytest -q -s \
+  tests/e2e/models/qwen3_moe/test_qwen3_moe.py
+~~~
+
 Do not add backend markers or run scenarios in parallel; they share devices.
 
-For the opt-in GPU/NPU 2A2F case, set four unique device IDs and run:
+For the opt-in DeepSeek-V2-Lite GPU/NPU 2A2F case, set four unique device IDs
+and run:
 
 ~~~bash
 export AFD_E2E_DEVICES=0,1,2,3
@@ -103,8 +115,9 @@ On cancellation, forward SIGTERM and allow over 90 seconds for cleanup.
 
 ### 5. Report
 
-Success means 4 passed and 0 skipped. Report the failed scenario, first
-actionable error, and cleanup status. Any skip is a gate failure.
+Success means the selected suite reports 4 passed and 0 skipped. Report the
+failed scenario, first actionable error, and cleanup status. Any skip is a
+gate failure.
 
 ## Environment reference
 
@@ -112,7 +125,7 @@ actionable error, and cleanup status. Any skip is a gate failure.
 |---|---|---|
 | AFD_E2E_BACKEND | both | yes: gpu or npu |
 | AFD_E2E_DEVICES | both | yes: exactly 3 unique IDs |
-| AFD_GPU_E2E_MODEL | GPU | no; downloads DeepSeek-V2-Lite when unset |
+| AFD_GPU_E2E_MODEL | GPU | no; downloads the selected suite's model when unset |
 | AFD_GPU_E2E_VLLM_BIN | GPU | no; defaults to vllm |
 | AFD_NPU_E2E_MODEL | NPU | no; downloads DeepSeek-V2-Lite when unset |
 | AFD_NPU_E2E_VLLM_BIN | NPU | no; defaults to vllm |
diff --git a/docs/design/module/e2e_testing.md b/docs/design/module/e2e_testing.md
index 2323e3ec..3e3247f6 100644
--- a/docs/design/module/e2e_testing.md
+++ b/docs/design/module/e2e_testing.md
@@ -25,6 +25,7 @@ validation_paths:
   - "tests/unit/test_e2e_process_utils.py"
   - "tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py"
   - "tests/e2e/models/deepseek_v2_lite/test_async_cam_npu.py"
+  - "tests/e2e/models/qwen3_moe/test_qwen3_moe.py"
 upstream_refs:
   - "vLLM 0.26.0 serving and shutdown interfaces"
   - "lm-evaluation-harness GSM8K task and local-completions API"
@@ -32,8 +33,9 @@ upstream_refs:
 verified_platform_refs:
   - "CUDA DeepSeek-V2-Lite"
   - "Ascend NPU DeepSeek-V2-Lite"
+  - "CUDA Qwen3 MoE"
 related_issues: []
-last_reviewed: 2026-08-12
+last_reviewed: 2026-08-13
 ---
 
 # E2E testing
@@ -139,5 +141,5 @@ of 1319.
 7. Update `tests/e2e/README.md`. Update the `run-e2e` skill only when commands,
    variables, prerequisites, or the default suite change.
 
-Before adding a second model, extract the shared environment, device, and
-runner-subprocess logic from `test_deepseek_v2_lite.py`.
+Model entrypoints reuse the subprocess and download helpers in
+`tests/conftest.py`.
diff --git a/tests/e2e/README.md b/tests/e2e/README.md
index fe9d88fa..f43eee6d 100644
--- a/tests/e2e/README.md
+++ b/tests/e2e/README.md
@@ -1,7 +1,7 @@
 # End-to-End Tests
 
-These tests validate DeepSeek-V2-Lite on real GPU or Ascend NPU hardware.
-The default gate runs four scenarios:
+These tests validate DeepSeek-V2-Lite on real GPU or Ascend NPU hardware and
+Qwen3 MoE on real GPU hardware. Each default gate runs four scenarios:
 
 - `baseline-graph`
 - `afd-eager`
@@ -27,9 +27,9 @@ Run from the repository root. The environment needs `vllm`, `pytest`,
 `afd_plugin`, `lm_eval`, `datasets`, and `huggingface_hub`. NPU also needs
 `torch_npu`.
 
-The test downloads/caches `openai/gsm8k` and
-`deepseek-ai/DeepSeek-V2-Lite` when the backend model env var is unset. Point
-`HF_HOME` at a persistent cache if you want to reuse downloads across runs.
+The selected test downloads/caches `openai/gsm8k` and its Hugging Face model
+when the backend model env var is unset. Point `HF_HOME` at a persistent cache
+if you want to reuse downloads across runs.
 
 GPU:
 
@@ -38,7 +38,7 @@ export HF_HOME=/path/to/huggingface
 export AFD_E2E_BACKEND=gpu
 # Optional: export AFD_E2E_DEVICES=0,1,2
 # Optional if the model is already local:
-# export AFD_GPU_E2E_MODEL=/path/to/DeepSeek-V2-Lite
+# export AFD_GPU_E2E_MODEL=/path/to/model
 ```
 
 NPU:
@@ -51,14 +51,19 @@ export AFD_E2E_BACKEND=npu
 # export AFD_NPU_E2E_MODEL=/path/to/DeepSeek-V2-Lite
 ```
 
-Then run:
+Then run the selected model suite:
 
 ```bash
+# DeepSeek-V2-Lite
 python -m pytest -q -s \
   "tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py::test_deepseek_v2_lite[afd-eager]" \
   "tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py::test_deepseek_v2_lite[afd-graph]" \
   "tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py::test_deepseek_v2_lite[afd-graph-dbo]" \
   "tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py::test_deepseek_v2_lite[baseline-graph]"
+
+# Qwen3 MoE
+python -m pytest -q -s \
+  tests/e2e/models/qwen3_moe/test_qwen3_moe.py
 ```
 
 Success means 4 passed and 0 skipped.
@@ -78,8 +83,13 @@ For the weekly full GSM8K test, run only `afd-graph-dbo`:
 
 ```bash
 export AFD_GSM8K_LIMIT=all
+# DeepSeek-V2-Lite
 python -m pytest -q -s \
   "tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py::test_deepseek_v2_lite[afd-graph-dbo]"
+
+# Qwen3 MoE
+python -m pytest -q -s \
+  "tests/e2e/models/qwen3_moe/test_qwen3_moe.py::test_qwen3_moe[afd-graph-dbo]"
 ```
 
 This evaluates all 1319 GSM8K test samples. Without `AFD_GSM8K_LIMIT`, each
@@ -91,7 +101,8 @@ The repository includes the [`run-e2e`](../../.agents/skills/run-e2e/SKILL.md)
 skill. Open the repository in Codex and ask, for example:
 
 ```text
-Use run-e2e to run the GPU E2E tests with HF_HOME /data/huggingface.
+Use run-e2e to run the Qwen3 MoE GPU E2E tests with HF_HOME
+/data/huggingface.
 ```
 
 Provide `HF_HOME` and `AFD_E2E_BACKEND`. `AFD_E2E_DEVICES` is optional; when
diff --git a/tests/e2e/models/qwen3_moe/test_qwen3_moe.py b/tests/e2e/models/qwen3_moe/test_qwen3_moe.py
new file mode 100644
index 00000000..131c0f7e
--- /dev/null
+++ b/tests/e2e/models/qwen3_moe/test_qwen3_moe.py
@@ -0,0 +1,105 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the AFD plugin project
+"""CUDA Qwen3 MoE E2E scenarios."""
+
+from __future__ import annotations
+
+import os
+import sys
+from pathlib import Path
+
+import pytest
+
+from tests.conftest import download_dataset, download_model, run_runner
+
+GSM8K_DATASET_ID = "openai/gsm8k"
+GSM8K_DATASET_CONFIG = "main"
+QWEN3_MOE_REPO_ID = "Qwen/Qwen3-30B-A3B"
+QWEN3_MOE_MAX_MODEL_LEN = 4096
+SCENARIOS = (
+    "baseline-graph",
+    "afd-eager",
+    "afd-graph",
+    "afd-graph-dbo",
+)
+
+
+def _required_env(name: str) -> str:
+    value = os.environ.get(name)
+    if not value:
+        raise RuntimeError(f"{name} must be set")
+    return value
+
+
+def prepare_e2e_assets() -> None:
+    """Ensure GSM8K and Qwen3-30B-A3B are available for the runner."""
+    download_dataset(GSM8K_DATASET_ID, GSM8K_DATASET_CONFIG)
+
+    if _required_env("AFD_E2E_BACKEND") != "gpu":
+        raise RuntimeError("Qwen3 MoE E2E supports only the 'gpu' backend")
+
+    existing = os.environ.get("AFD_GPU_E2E_MODEL")
+    if existing:
+        print(
+            f"[e2e] Using existing model path {Path(existing).expanduser()}",
+            flush=True,
+        )
+        return
+
+    os.environ["AFD_GPU_E2E_MODEL"] = str(download_model(QWEN3_MOE_REPO_ID))
+
+
+def build_runner_command(scenario: str, gsm8k_output_path: Path) -> list[str]:
+    backend = _required_env("AFD_E2E_BACKEND")
+    if backend != "gpu":
+        raise RuntimeError("Qwen3 MoE E2E supports only the 'gpu' backend")
+
+    env_devices = os.environ.get("AFD_E2E_DEVICES")
+    devices = (
+        [item.strip() for item in env_devices.split(",") if item.strip()]
+        if env_devices
+        else ["0", "1", "2"]
+    )
+    attention_devices = devices[:2]
+    ffn_devices = devices[2:]
+
+    command = [
+        sys.executable,
+        "-m",
+        "tests.e2e.runner",
+        "--model",
+        _required_env("AFD_GPU_E2E_MODEL"),
+        "--vllm-bin",
+        os.environ.get("AFD_GPU_E2E_VLLM_BIN", "vllm"),
+        "--device-backend",
+        backend,
+        f"--common-vllm-arg=--max-model-len={QWEN3_MOE_MAX_MODEL_LEN}",
+        "--attention-devices",
+        ",".join(attention_devices),
+    ]
+    if scenario != "baseline-graph":
+        command.extend(["--ffn-devices", ",".join(ffn_devices)])
+    command.extend(
+        [
+            "--scenario",
+            scenario,
+            "--gsm8k-output-path",
+            str(gsm8k_output_path),
+            "--served-model-name-prefix",
+            "qwen3-moe-afd",
+        ],
+    )
+    return command
+
+
+@pytest.fixture(scope="module", autouse=True)
+def _prepare_e2e_assets() -> None:
+    """Prepare shared E2E assets once for this test module."""
+    prepare_e2e_assets()
+
+
+@pytest.mark.e2e
+@pytest.mark.parametrize("scenario", SCENARIOS, ids=SCENARIOS)
+def test_qwen3_moe(scenario: str, tmp_path: Path) -> None:
+    command = build_runner_command(scenario, tmp_path / scenario)
+    run_runner(command)
diff --git a/tests/unit/test_e2e_runner.py b/tests/unit/test_e2e_runner.py
index 3010fc5d..63d5093d 100644
--- a/tests/unit/test_e2e_runner.py
+++ b/tests/unit/test_e2e_runner.py
@@ -16,6 +16,7 @@
 from tests.e2e.models.deepseek_v2_lite import (
     test_deepseek_v2_lite as deepseek_v2_lite_e2e,
 )
+from tests.e2e.models.qwen3_moe import test_qwen3_moe as qwen3_moe_e2e
 
 
 def test_baseline_entrypoint_uses_two_devices(monkeypatch, tmp_path):
@@ -47,6 +48,28 @@ def test_deepseek_v2_lite_entrypoint_limits_context(
     assert "--common-vllm-arg=--max-model-len=4096" in command
 
 
+def test_qwen3_moe_baseline_entrypoint_uses_two_devices(monkeypatch, tmp_path):
+    monkeypatch.setenv("AFD_E2E_BACKEND", "gpu")
+    monkeypatch.setenv("AFD_E2E_DEVICES", "2,4,6")
+    monkeypatch.setenv("AFD_GPU_E2E_MODEL", "model")
+
+    command = qwen3_moe_e2e.build_runner_command("baseline-graph", tmp_path)
+
+    assert command[command.index("--attention-devices") + 1] == "2,4"
+    assert "--ffn-devices" not in command
+
+
+@pytest.mark.parametrize("scenario", qwen3_moe_e2e.SCENARIOS)
+def test_qwen3_moe_entrypoint_limits_context(monkeypatch, tmp_path, scenario):
+    monkeypatch.setenv("AFD_E2E_BACKEND", "gpu")
+    monkeypatch.setenv("AFD_GPU_E2E_MODEL", "model")
+    monkeypatch.delenv("AFD_E2E_DEVICES", raising=False)
+
+    command = qwen3_moe_e2e.build_runner_command(scenario, tmp_path)
+
+    assert "--common-vllm-arg=--max-model-len=4096" in command
+
+
 def test_run_runner_forwards_cancellation_and_reaps(monkeypatch):
     command = [sys.executable, "-m", "tests.e2e.runner"]
     previous_handlers = {