Qwen2.5-Omni

What it is

Qwen2.5-Omni is a multimodal model from Alibaba's Qwen team. It perceives text, images, audio and video, and responds with both text and natural speech. It is described as a single end-to-end model.^[1]

Origins

The 7B version, Qwen2.5-Omni-7B, was released on 26 March 2025 under Apache 2.0. A 3B version was released under the Qwen Research license.^[2] The technical report was posted to arXiv on the same day.^[3]

The model grew out of Alibaba's earlier audio work. Qwen-Audio arrived in November 2023 and was pre-trained on more than 30 tasks covering speech, natural sounds, music and songs. Qwen2-Audio followed in August 2024.^[4][2]

Main features

The model uses a Thinker-Talker architecture. The Thinker produces text, and the Talker turns that text into streaming speech.

The team introduced Time-aligned Multimodal RoPE (TMRoPE), a position encoding that synchronises video timestamps with audio.^[1]

The model is designed for real-time interaction, with chunked input and immediate output. It offers two voices, Chelsie and Ethan.^[1]

On OmniBench, the model averaged 56.13%. Gemini-1.5-Pro averaged 42.91% on the same test.^[1] On vision tasks it matched Qwen2.5-VL-7B, and on audio tasks it outperformed Qwen2-Audio.^[1]

Several terms help explain the design. End-to-end means that one model handles listening, understanding and speaking, rather than passing the job between separate systems. Multimodal means the model accepts more than one kind of input. Streaming speech is audio that starts playing while the rest of the answer is still being generated. OmniBench is a test designed to check how well a model combines text, images and audio when answering questions.^[1]

Licensing matters for reuse. Apache 2.0 is a permissive open-source license that allows use and modification. The 3B version was released under the Qwen Research license, and users should read the terms of whichever version they use.^[2]

Place in development

Qwen2.5-Omni was succeeded by Qwen3-Omni in September 2025.^[2]