diff --git a/baselines/main.py b/baselines/main.py
index 56d73ff..2302c73 100644
--- a/baselines/main.py
+++ b/baselines/main.py
@@ -31,6 +31,8 @@
 from models.mllm.ola import load_model as load_ola
 from models.mllm.qwen_omni import generate as generate_qwen_omni
 from models.mllm.qwen_omni import load_model as load_qwen_omni
+from models.mllm.qwen3_omni import generate as generate_qwen3_omni
+from models.mllm.qwen3_omni import load_model as load_qwen3_omni
 
 # import speech models
 from models.speech.desta import generate as generate_desta
@@ -141,6 +143,9 @@ def load_model(model_name):
     elif model_name == "qwen_omni":
         model = load_qwen_omni()
         generate_func = generate_qwen_omni
+    elif model_name == "qwen3_omni":
+        model = load_qwen3_omni()
+        generate_func = generate_qwen3_omni
     elif model_name == "ming_lite_omni":
         model = load_ming_lite_omni()
         generate_func = generate_ming_lite_omni
@@ -228,6 +233,7 @@ def main(in_data_folder, out_folder, model, lang, track, modality, prompt):
         "ming_lite_omni",
         "ola",
         "qwen_omni",
+        "qwen3_omni",
         "gpt_oss",
         "gemini",
     ]
diff --git a/baselines/models/mllm/qwen3_omni.py b/baselines/models/mllm/qwen3_omni.py
new file mode 100644
index 0000000..a5e0ec1
--- /dev/null
+++ b/baselines/models/mllm/qwen3_omni.py
@@ -0,0 +1,118 @@
+# Copyright 2026 FBK, KIT
+
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+
+#     http://www.apache.org/licenses/LICENSE-2.0
+
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License
+
+from utils import read_txt_file
+
+
+def load_model():
+    from transformers import Qwen3OmniMoeForConditionalGeneration, Qwen3OmniMoeProcessor
+
+    model = Qwen3OmniMoeForConditionalGeneration.from_pretrained(
+        "Qwen/Qwen3-Omni-30B-A3B-Instruct",
+        torch_dtype="auto",
+        device_map="auto",
+        attn_implementation="flash_attention_2",
+    )
+    processor = Qwen3OmniMoeProcessor.from_pretrained("Qwen/Qwen3-Omni-30B-A3B-Instruct")
+
+    return model, processor
+
+
+def generate(model_processor, prompt, example_path, modality):
+    from qwen_omni_utils import process_mm_info
+
+    model, processor = model_processor
+
+    if modality == "mllm":
+        USE_AUDIO_IN_VIDEO = True
+        user_conv_content = [
+            {"type": "video", "video": example_path},
+            {"type": "text", "text": prompt},
+        ]
+
+    elif modality == "video":
+        USE_AUDIO_IN_VIDEO = False
+        user_conv_content = [
+            {"type": "video", "video": example_path},
+            {"type": "text", "text": prompt},
+        ]
+
+    elif modality == "audio":
+        USE_AUDIO_IN_VIDEO = False
+        user_conv_content = [
+            {"type": "audio", "audio": example_path},
+            {"type": "text", "text": prompt},
+        ]
+
+    elif modality == "text":
+        USE_AUDIO_IN_VIDEO = False
+        example = read_txt_file(example_path)
+        user_conv_content = [
+            {"type": "text", "text": f"{example}\n{prompt}\n"},
+        ]
+
+    system_conv = {
+        "role": "system",
+        "content": [
+            {
+                "type": "text",
+                "text": "You are Qwen, a virtual human developed by the Qwen Team, Alibaba Group, "
+                        "capable of perceiving auditory and visual inputs, as well as generating "
+                        "text and speech. Only return the answer requested. Do not include any "
+                        "explanation or introductions.",
+            }
+        ],
+    }
+
+    user_conv = {
+        "role": "user",
+        "content": user_conv_content,
+    }
+
+    conversation = [system_conv, user_conv]
+
+    # Preparation for inference
+    text = processor.apply_chat_template(
+        conversation, add_generation_prompt=True, tokenize=False
+    )
+    audios, images, videos = process_mm_info(
+        conversation, use_audio_in_video=USE_AUDIO_IN_VIDEO
+    )
+    inputs = processor(
+        text=text,
+        audio=audios,
+        images=images,
+        videos=videos,
+        return_tensors="pt",
+        padding=True,
+        use_audio_in_video=USE_AUDIO_IN_VIDEO,
+    )
+    inputs = inputs.to(model.device).to(model.dtype)
+
+    # Qwen3-Omni always returns a (text_ids, audio) tuple; with return_audio=False the
+    # second element is None. thinker_max_new_tokens/thinker_do_sample are Qwen2.5-Omni
+    # specific parameters and are not supported by Qwen3-Omni.
+    text_ids, _ = model.generate(
+        **inputs,
+        use_audio_in_video=USE_AUDIO_IN_VIDEO,
+        return_audio=False,
+        max_new_tokens=4096,
+    )
+    text = processor.batch_decode(
+        text_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
+    )
+
+    # postprocess
+    response = text[-1].split("\nassistant")[-1].strip()
+    return response