diff --git a/env_files/glm_4d1v_requirements.txt b/env_files/glm_4d1v_requirements.txt
new file mode 100644
index 00000000..6004f8af
--- /dev/null
+++ b/env_files/glm_4d1v_requirements.txt
@@ -0,0 +1,12 @@
+torch==2.6.0
+torchvision==0.21.0
+transformers==4.57.1
+accelerate
+datasets
+pandas
+numpy==1.26.4
+Pillow
+sentencepiece
+protobuf
+einops
+timm
diff --git a/mmeval/infer/glm_4d1v.py b/mmeval/infer/glm_4d1v.py
new file mode 100644
index 00000000..6f666adf
--- /dev/null
+++ b/mmeval/infer/glm_4d1v.py
@@ -0,0 +1,96 @@
+"""glm_4d1v — GLM-4.1V-9B-Thinking.
+
+HF: https://huggingface.co/THUDM/GLM-4.1V-9B-Thinking
+GH: https://github.com/THUDM/GLM-V
+"""
+import re
+import copy
+
+import torch
+from PIL import Image
+from transformers import AutoProcessor, Glm4vForConditionalGeneration
+
+from mmeval.infer.task import Task
+from mmeval.utils import constants
+from mmeval.utils.argparser import parse_args, parse_model_kwargs, parse_gen_kwargs
+
+
+class TaskRunner(Task):
+    def __init__(self, args):
+        self.args = args
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        self.dtype = getattr(args, "dtype") or torch.bfloat16
+        self.default_model_kwargs = {"device_map": "auto"}
+        self.default_gen_kwargs = {"max_new_tokens": 8192}
+        self.model_kwargs = parse_model_kwargs(args, self.default_model_kwargs)
+        self.gen_kwargs = parse_gen_kwargs(args, self.default_gen_kwargs)
+
+        super().__init__(args)
+
+    def load_model(self, args):
+        self.model = Glm4vForConditionalGeneration.from_pretrained(
+            args.model_name_or_path,
+            torch_dtype=self.dtype,
+            **self.model_kwargs,
+        ).eval()
+        self.processor = AutoProcessor.from_pretrained(
+            args.model_name_or_path, use_fast=True,
+        )
+
+    def parse_input(self, message):
+        question = message["prompt"]
+        q_chunks = re.split(r'(<(?:image|video)>)', question)
+        media_list = message.get('media', [])
+
+        content = []
+        media_idx = 0
+        for chunk in q_chunks:
+            if not chunk.strip():
+                continue
+            if chunk == constants.image:
+                img = media_list[media_idx]
+                if isinstance(img, str):
+                    img = Image.open(img).convert("RGB")
+                elif hasattr(img, "convert"):
+                    img = img.convert("RGB")
+                content.append({"type": "image", "image": img})
+                media_idx += 1
+            elif chunk == constants.video:
+                content.append({"type": "video", "video": media_list[media_idx]})
+                media_idx += 1
+            else:
+                content.append({"type": "text", "text": chunk})
+        return [{"role": "user", "content": content}]
+
+    def _generate_response(self, inputs):
+        generated_ids = self.model.generate(**inputs, **self.gen_kwargs)
+        trimmed = generated_ids[0][inputs["input_ids"].shape[1]:]
+        return self.processor.decode(trimmed, skip_special_tokens=False)
+
+    def run_sample(self, sample: dict):
+        if self.args.score_target:
+            raise NotImplementedError(
+                "glm_4d1v: score_target is not implemented yet"
+            )
+
+        ori_sample = copy.deepcopy(sample)
+        message = sample["messages"][0]
+        messages = self.parse_input(message)
+
+        inputs = self.processor.apply_chat_template(
+            messages,
+            tokenize=True,
+            add_generation_prompt=True,
+            return_dict=True,
+            return_tensors="pt",
+        ).to(self.model.device)
+
+        response = self._generate_response(inputs)
+        ori_sample["messages"].append({"role": "assistant", "response": response})
+        return ori_sample
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    model_evaluator = TaskRunner(args)
+    model_evaluator.inference_dataset()
diff --git a/mmeval/registry.py b/mmeval/registry.py
index dfbf0b98..cc3160d2 100644
--- a/mmeval/registry.py
+++ b/mmeval/registry.py
@@ -77,6 +77,7 @@
                    "doubao-seed-2-0-mini-260215", "doubao-seed-2-0-lite-260215", "doubao-seed-2-0-code-preview-260215", "doubao-seed-2-0-pro-260215"],
     "hunyuan_vision": ["hunyuan-vision", "hunyuan-vision-1.5-instruct", "hunyuan-t1-vision", "hunyuan-turbos-vision", "hunyuan-large-vision"],
     "cosmos_reason2": ["Cosmos-Reason2-2B", "Cosmos-Reason2-8B"],
+    "glm_4d1v": ["GLM-4.1V-9B-Thinking", "GLM-4.1V-9B-Base"],
 }
 
 series_infer_env_mapping = {
@@ -316,4 +317,8 @@
         "env": os.path.join(env_dir, "cosmos_reason2"),
         "infer_file": "cosmos_reason2.py",
     },
+    "glm_4d1v": {
+        "env": os.path.join(env_dir, "glm_4d1v"),
+        "infer_file": "glm_4d1v.py",
+    },
 }
diff --git a/test_results/glm_4d1v_2026-06-10/no_media/result.json b/test_results/glm_4d1v_2026-06-10/no_media/result.json
new file mode 100644
index 00000000..040588f2
--- /dev/null
+++ b/test_results/glm_4d1v_2026-06-10/no_media/result.json
@@ -0,0 +1,44 @@
+[
+    {
+        "id": 0,
+        "media": [],
+        "messages": [
+            {
+                "role": "user",
+                "question": "What is the result of 1 plus 1?",
+                "answer": "",
+                "options": {},
+                "choices": [],
+                "prompt": "What is the result of 1 plus 1?",
+                "hint": ""
+            },
+            {
+                "role": "assistant",
+                "response": "<think>Got it, let's see. The question is asking for the result of 1 plus 1. That's a basic arithmetic operation. When you add"
+            }
+        ],
+        "comment": "Test 'no media' modality.",
+        "eval-id": 0
+    },
+    {
+        "id": 1,
+        "media": [],
+        "messages": [
+            {
+                "role": "user",
+                "question": "What is the capital city of China?",
+                "answer": "",
+                "options": {},
+                "choices": [],
+                "prompt": "What is the capital city of China?",
+                "hint": ""
+            },
+            {
+                "role": "assistant",
+                "response": "<think>Got it, the question is asking for the capital city of China. I need to recall that the capital of China is Beijing. Let me make sure there"
+            }
+        ],
+        "comment": "Test 'no media' modality.",
+        "eval-id": 1
+    }
+]
\ No newline at end of file
diff --git a/test_results/glm_4d1v_2026-06-10/single_image_start/result.json b/test_results/glm_4d1v_2026-06-10/single_image_start/result.json
new file mode 100644
index 00000000..544174c4
--- /dev/null
+++ b/test_results/glm_4d1v_2026-06-10/single_image_start/result.json
@@ -0,0 +1,48 @@
+[
+    {
+        "id": 0,
+        "media": [
+            "truck.png"
+        ],
+        "messages": [
+            {
+                "role": "user",
+                "question": "<image> Please provide a detailed description of the contents shown in the image.",
+                "answer": "",
+                "options": {},
+                "choices": [],
+                "prompt": "<image> Please provide a detailed description of the contents shown in the image.",
+                "hint": ""
+            },
+            {
+                "role": "assistant",
+                "response": "<think>Got it, let's start by describing the image. The main subject is a GMC pickup truck, specifically a modern model, probably the 2024 GMC Hummer EV or a similar model, but wait, no, the design looks like the GMC Sierra or maybe the Hummer pickup? Wait, the image shows"
+            }
+        ],
+        "comment": "Test 'single image start' modality.",
+        "eval-id": 0
+    },
+    {
+        "id": 1,
+        "media": [
+            "boat.png"
+        ],
+        "messages": [
+            {
+                "role": "user",
+                "question": "<image> Please provide a detailed description of the contents shown in the image.",
+                "answer": "",
+                "options": {},
+                "choices": [],
+                "prompt": "<image> Please provide a detailed description of the contents shown in the image.",
+                "hint": ""
+            },
+            {
+                "role": "assistant",
+                "response": "<think>Got it, let's analyze the image. The main subject is a large white yacht or motorboat moving through the water. The boat is sleek, with a modern design, probably a luxury yacht. It's creating a big wake as it moves, which shows it's going at a good speed. The water is"
+            }
+        ],
+        "comment": "Test 'single image start' modality.",
+        "eval-id": 1
+    }
+]
\ No newline at end of file
diff --git a/test_results/glm_4d1v_2026-06-10/test_summary.json b/test_results/glm_4d1v_2026-06-10/test_summary.json
new file mode 100644
index 00000000..f9e85804
--- /dev/null
+++ b/test_results/glm_4d1v_2026-06-10/test_summary.json
@@ -0,0 +1,20 @@
+{
+  "model_series": "glm_4d1v",
+  "model_name": "THUDM/GLM-4.1V-9B-Thinking",
+  "test_date": "2026-06-10",
+  "conda_env": "/raid/ztw/envs/glm_4d1v",
+  "env_summary": "transformers==4.57.1",
+  "status": "PASS",
+  "modalities_tested": {
+    "no_media": {
+      "status": "pass",
+      "rows": 2,
+      "result_path": "/raid/ztw/simple-mmeval-test-result/work_dirs/glm_4d1v/2026-06-10/no_media/result.json"
+    },
+    "single_image_start": {
+      "status": "pass",
+      "rows": 2,
+      "result_path": "/raid/ztw/simple-mmeval-test-result/work_dirs/glm_4d1v/2026-06-10/single_image_start/result.json"
+    }
+  }
+}
\ No newline at end of file
diff --git a/test_results/test_glm_4d1v.sh b/test_results/test_glm_4d1v.sh
new file mode 100755
index 00000000..1cb90b67
--- /dev/null
+++ b/test_results/test_glm_4d1v.sh
@@ -0,0 +1,25 @@
+#!/usr/bin/env bash
+# Smoke test for glm_4d1v (model: THUDM/GLM-4.1V-9B-Thinking).
+set -euo pipefail
+
+REPO_DIR=${REPO_DIR:-/raid/ztw/simple-mmeval-skills-dev}
+OUT_ROOT=${OUT_ROOT:-/raid/ztw/simple-mmeval-test-result/work_dirs/glm_4d1v/$(date +%Y-%m-%d)}
+GPU=${GPU:-0}
+MODEL_ID=THUDM/GLM-4.1V-9B-Thinking
+
+cd "$REPO_DIR"
+mkdir -p "$OUT_ROOT"
+
+for spec in "no_media|32|" "single_image_start|64|tests/media/448"; do
+  IFS='|' read -r sample maxtok imgdir <<< "$spec"
+  out="$OUT_ROOT/$sample"
+  extra=()
+  [[ -n "$imgdir" ]] && extra=(--img_dir "$imgdir")
+  PYTHONPATH="$REPO_DIR" ENV_DIR=/raid/ztw/envs CUDA_VISIBLE_DEVICES="$GPU" \
+    /raid/ztw/envs/glm_4d1v/bin/python mmeval/run.py \
+      --model_name_or_path "$MODEL_ID" \
+      --dataset local@json --infile tests/samples/$sample.json \
+      --out_dir "$out" \
+      --gpu_per_parallel 1 --parallel_per_task 1 --max_new_tokens "$maxtok" \
+      "${extra[@]}"
+done