Alibaba Unveils Qwen2.5-Omni-7B: A Breakthrough in Open-Source Multimodal AI

Alibaba launches new open-source AI model for 'cost-effective AI agents'

The Alibaba office building in Nanjing, Jiangsu province, China, on Aug 28, 2024. Alibaba Cloud launched Thursday its latest AI model in its "Qwen series," as large language model competition in China continues to heat up following the "DeepSeek moment." The new "Qwen2.5-Omni-7B" is a multimodal model, which means it can process inputs, including text, images, audio and videos, while generating real-time text and natural speech responses, according to an announcement on Alibaba Cloud's website. The company says that the model can be deployed on edge devices like mobile phones, offering high efficiency without compromising performance. "This unique combination makes it the perfect foundation for developing agile, cost-effective AI agents that deliver tangible value, especially intelligent voice applications," Alibaba said. For example, it could be used to help a visually impaired person navigate their environment through real-time audio description, the company added. The new model is open-sourced on the platforms Hugging Face and Github, following a growing trend in China after DeepSeek made its breakthrough R1 model open-source. Open-source generally refers to software in which the source code is made freely available on the web for possible modification and redistribution. Over the past years, Alibaba Cloud says it has open-sourced over 200 generative AI models. Amid China's AI fervor accelerated by DeepSeek, Alibaba and other generative AI competitors have been releasing new, cost-effective models and products at an unprecedented pace. Last week, Chinese tech giant Baidu released a new multimodal foundational model and its first reasoning-focused model. Alibaba, meanwhile, debuted its updated Qwen 2.5 artificial intelligence model in late January and released a new version of its AI assistant tool Quark earlier this month. The company has strongly committed to its AI strategy, announcing last month a plan to invest $53 billion in its cloud computing and AI infrastructure over the next three years, exceeding what it spent in the space over the past decade. Kai Wang, Asia senior equity analyst at Morningstar, told CNBC that large Chinese tech players such as Alibaba, which build data centers to meet the computing needs of AI in addition to building their own LLMs, are well positioned to benefit from China's post-DeepSeek AI boom.

Alibaba Releases Qwen2.5 Omni, Adds Voice and Video Modes to Qwen Chat

Alibaba's Qwen2.5-Omni model has shown strong performance across various tasks, including speech recognition, audio, video, and more. Alibaba, on Wednesday, added voice and video chat capabilities to Qwen Chat, besides releasing its brand new open-source model, Qwen2.5-Omni-7B, which made this possible. It was released as an open-source model under Apache 2.0 licence. The company highlighted in a blog post that Qwen2.5-Omni is the new flagship end-to-end multimodal model in the Qwen series. It stated that it is designed for multimodal perception and seamlessly processes text, images, audio, and video, delivering real-time streaming responses via text and speech synthesis. The key features of the model include a 'Thinker-Talker' architecture, which allows it to provide real-time responses. The Thinker part of the architecture is a Transformer decoder, which acts like the brain and the Talker, designed as a dual-track autoregressive Transformer decoder, operates like the human mouth. Alibaba's Qwen2.5-Omni model has shown strong performance across various tasks, including speech recognition, translation, audio and video understanding, and speech generation, outperforming similar models at tasks that require multiple modalities. It was compared to similar single-modality and closed-source models like Qwen2.5-VL-7B, Qwen2-Audio, and Gemini-1.5-pro, achieving state-of-the-art performance. The paper and code for the new model can be found on GitHub, while the AI model is available on Hugging Face along with a demo. Last month, Alibaba also launched QwQ-Max-Preview, a new AI reasoning model within the Qwen family that specialises in mathematics and coding tasks and features a "thinking" capability in the Qwen Chat application. The model, which outperformed OpenAI's models on the LiveCodeBench leaderboard, is expected to have smaller variants open-sourced for local device deployment, as well as a dedicated mobile app. There may be a lot more coming, considering Alibaba's commitment to investing over $52 billion in AI over the next three years.

Alibaba releases new open-source AI model to power intelligent voice applications - SiliconANGLE

The company said the model, which is named Qwen2.5-Omni-7B, is small enough that it will fit on devices such as mobile phones and similar devices. Despite its compact size, at only 7 billion parameters, Alibaba Cloud said it provides high performance and powerful multimodal capabilities. It is capable of understanding video inputs from cameras and watching the screen as the user operates the device to respond in real time. This means that it can be combined with applications to hold conversations. "This unique combination makes it the perfect foundation for developing agile, cost-effective AI agents that deliver tangible value, especially intelligent voice applications," the company said in the announcement. Users could use the model to provide real-time assistance while shopping, step-by-step cooking guidance by analyzing video ingredients, or even read through a PDF on the screen to assist with tedious research. The video capabilities of the model could also make it ideal for visually impaired users to navigate environments because it can read signs, understand context clues and match voices to faces. The company released the model open-source on Hugging Face and GitHub. It is additionally accessible on Qwen Chat and through the company's open-source community ModelScope. Open source refers to a type of software development where the code and weights of the AI models are freely available for developers to use, modify and distribute. This community-centric model promotes collaboration and Alibaba Cloud has released over 200 generative AI models open source to date. Since the open-source release of DeepSeek-R1, from the China-based AI developer of the same name, Chinese companies have been making headway in the AI market with significant model releases. DeepSeek's R1 model family introduced reasoning capabilities where models could "think" through problems, and last month Chinese technology giant Tencent Holdings Ltd. released Hunyuan Turbo S, which the company claimed outperformed R1. Last week, Chinese multinational internet search giant Baidu released a multimodal foundational model and its first reasoning-focused model Ernie-X1 to compete with DeepSeek.

Alibaba preparing for flagship AI model release as soon as April

Gift 5 articles to anyone you choose each month when you subscribe. Alibaba Group Holding is planning to release Qwen 3, an upgraded version of its flagship AI model, as soon as this month with competition from rivals including OpenAI and DeepSeek heating up. The Hangzhou-based company's new offering may arrive later in April, though the exact timing could still slip, a person familiar with the matter said, asking not to be identified because the information isn't public. The Chinese media outlet Huxiu earlier reported about Alibaba's plans.

Alibaba Qwen 2.5 VL AI Model Released; Now in a 32 Billion Parameter Size

Alibaba says its responses are more aligned with human preferences Alibaba's Qwen team released another artificial intelligence (AI) model to the Qwen 2.5 family on Monday. Dubbed Qwen 2.5-VL-32B Instruct, the AI model comes with improved performance and optimisations. It is a vision language model with 32 billion parameters, and joins the three billion, seven billion, and 72 billion parameter size models in the Qwen 2.5 family. Just like all previous models by the team, it is also an open-source AI model available under a permissive license. In a blog post, the Qwen team detailed the company's latest vision language model (VLM). It is more capable than the Qwen 2.5 3B and 7B models, and smaller than the foundation 72B model. The large language model's (LLM) older versions outperformed DeepSeek-V3, and the 32B model is said to be outperforming Google and Mistral's similar sized systems. Coming to its features, the Qwen 2.5-VL-32B-Instruct has an adjusted output style that provides more detailed and better-formatted responses. The researchers claimed that the responses are closely aligned with human preferences. Mathematical reasoning capability has also been improved, and the AI model can solve more complex problems. The accuracy of image understanding capability and reasoning-focused analysis, including image parsing, content recognition, and visual logic deduction, has also been improved. Based on internal testing, the Qwen 2.5-VL-32B is claimed to have surpassed the capabilities of comparable models, such as Mistral-Small-3.1-24B and Google's Gemma-3-27B, on the MMMU, MMMU-Pro, and MathVista benchmarks. Interestingly, the LLM was also claimed to have outperformed the much larger Qwen 2-VL-72B model on the MM-MT-Bench. The Qwen team highlights that the latest model can directly play as a visual agent that can reason and direct tools. It is inherently capable of computer use and phone use. It accepts text, images, and videos with more than one hour of duration as input. It also supports JSON and structured outputs. The baseline architecture and training remain the same as the older Qwen 2.5 models, however, the researchers implemented a dynamic fps sampling to enable the model to comprehend videos at varying sampling rates. Another enhancement also lets it pinpoint specific moments in a video by gaining an understanding of temporal sequence and speed. Qwen 2.5-VL-32B-Instruct is available to download on GitHub and its Hugging Face listing. The model comes with Apache 2.0 licence, which allows both academic and commercial usage.

Alibaba's Qwen 2.5 Omni AI Model to Help Develop Cost-Effective AI Agents

Alibaba said the AI model uses the Thinker-Talker architecture Alibaba's Qwen team released a new artificial intelligence (AI) model in the Qwen 2.5 family on Wednesday. Dubbed Qwen 2.5 Omni, it is a flagship-tier end-to-end multimodal model. The company claims it can process a wide range of inputs, including text, images, audio, and videos, while generating real-time text and natural speech responses. It is said to enable the building and deployment of cost-effective AI agents due to its diverse skill set. Alibaba has also employed a new "Thinker-Talker" architecture for the Qwen 2.5 Omni AI model. In a blog post, the Qwen team detailed the new Qwen 2.5 Omni AI model, which is a seven-billion-parameter system. The most notable capability of this omnimodal model is the real-time speech generation and video chat capability, which will allow the large language model (LLM) to answer queries and interact with users verbally in a humanlike manner. So far, this capability is only available with Google and OpenAI's models, which are closed-source. Alibaba, on the other hand, has open-sourced the technology. Coming to the features, it accepts text, images, audio, and video as input as well as output. The model is also capable of real-time voice interactions and video chats. The Qwen team also highlights that the model will also offer real-time streaming of speech in a natural manner. Additionally, it is claimed to come with enhanced performance in end-to-end speech instruction. The Qwen team highlighted that the Omni model is built on a novel "Thinker-Talker" architecture. The Thinker component functions like a brain and is responsible for processing and understanding input across modalities, and generating text output. It is essentially a Transformer decoder that encodes audio and image and assists with information extraction. On the other hand, the Talker component operates like a human mouth, the researchers said. It streams the information produced by the Thinker component and generates a stream-like output for speech fluidity. It is designed as a dual-track autoregressive Transformer decoder. This entire architecture operates as a single model, allowing real-time text and speech generation, enabling end-to-end training and inference. Based on internal testing, the Qwen 2.5 Omni AI model is said to outperform the Gemini 1.5 Pro model on the OmniBench. It also outperforms Qwen 2.5-VL-7B, Qwen2-Audio on single-modality tasks. The AI model is now available on Alibaba's Hugging Face listing and GitHub listing. Additionally, users can test out the new model via Qwen Chat as well as the company's community ModelScope.

Alibaba Preparing for Flagship AI Model Release as Soon as April

China's tech leaders have flooded the market with low cost AI services Alibaba Group Holding is planning to release Qwen 3, an upgraded version of its flagship AI model, as soon as this month with competition from rivals including OpenAI and DeepSeek heating up. The Hangzhou-based company's new offering may arrive later in April, though the exact timing could still slip, a person familiar with the matter said, asking not to be identified because the information isn't public. The Chinese media outlet Huxiu earlier reported about Alibaba's plans. Alibaba has been releasing AI products at a frenetic pace since going all-in on the technology this year. The ecommerce and cloud computing leader in China came out with a new model in its Qwen 2.5 series just a week ago that can process text, pictures, audio and video -- and is efficient enough to run directly on mobile phones and laptops. Last month, it also unveiled a new version of the AI assistant Quark app. Since Alibaba's Hangzhou peer DeepSeek upstaged OpenAI with a powerful model that purportedly cost just several million dollars to build, China's tech leaders have flooded the market with a rapid succession of low-cost AI services. The wave of new models out of Asia are threatening to undercut premium US offerings from the likes of OpenAI, Alphabet's Google and Microsoft. OpenAI, Google and Anthropic have similarly released a flurry of new models in recent weeks. OpenAI recently said it also plans to release a more "open" model that mimics human reasoning in the coming months, a shift in strategy after DeepSeek and Alibaba pushed out open-source AI systems.

Alibaba preparing for flagship AI model release as soon as April

Alibaba Group plans to launch Qwen 3, an upgraded AI model, possibly in April, amid rising competition from OpenAI and DeepSeek. The move follows Alibaba's rapid AI developments, including the Qwen 2.5 and Quark app, as Chinese firms challenge US giants with cost-effective AI solutions.Alibaba Group is planning to release Qwen 3, an upgraded version of its flagship AI model, as soon as this month with competition from rivals including OpenAI and DeepSeek heating up. The Hangzhou-based company's new offering may arrive later in April, though the exact timing could still slip, a person familiar with the matter said, asking not to be identified because the information isn't public. The Chinese media outlet Huxiu earlier reported about Alibaba's plans. Alibaba has been releasing AI products at a frenetic pace since going all-in on the technology this year. The ecommerce and cloud computing leader in China came out with a new model in its Qwen 2.5 series just a week ago that can process text, pictures, audio and video -- and is efficient enough to run directly on mobile phones and laptops. Last month, it also unveiled a new version of the AI assistant Quark app. Since Alibaba's Hangzhou peer DeepSeek upstaged OpenAI with a powerful model that purportedly cost just several million dollars to build, China's tech leaders have flooded the market with a rapid succession of low-cost AI services. The wave of new models out of Asia are threatening to undercut premium US offerings from the likes of OpenAI, Alphabet's Google and Microsoft. OpenAI, Google and Anthropic have similarly released a flurry of new models in recent weeks. OpenAI recently said it also plans to release a more "open" model that mimics human reasoning in the coming months, a shift in strategy after DeepSeek and Alibaba pushed out open-source AI systems.

Alibaba Cloud Launches Compact, Multimodal AI Model | PYMNTS.com

Alibaba Cloud has launched a multimodal artificial intelligence (AI) model that can process inputs in the form of text, images, audio and video, and can generate real-time responses in the form of text and natural speech. The new Qwen2.5-Omni-7B can be deployed on mobile phones and laptops, the company said in an article posted on Alibaba's news website, Alizila. Because the model is both compact and multimodal, it can power "agile, cost-effective AI agents," according to the article. "For example, the model could be leveraged to transform lives by helping visually impaired users navigate environments through real-time audio descriptions, offering step-by-step cooking guidance by analyzing video ingredients, or powering intelligent customer service dialogues that really understand customer needs," the article said. Qwen2.5-Omni-7B is open-sourced on Hugging Face and GitHub and can be accessed via Qwen Chat and ModelScope, which is Alibaba Cloud's open-source community, per the article. Among the more than 200 generative AI models open-sourced by Alibaba Cloud, the new model stands apart in terms of its performance across all modalities and the "new benchmark" it set in real-time voice interaction, natural and robust speech generation, and following end-to-end speech instructions, the article said. This announcement came about two months after Alibaba released an AI model called Qwen2.5-Max and said it outperforms top AI models on key benchmarks. Alibaba said at the time that Qwen2.5-Max held its own against DeepSeek V3, Llama 3.1-405B, GPT-4o and Claude 3.5 Sonnet in the MMLU-Pro, GPQA-Diamond, LiveCodeBench, LiveBench and Arena-Hard benchmarks. In February, Alibaba said during an earnings call that it will spend more on AI in the next three years than it has in the last decade. "We aim to continue to develop models that extend the boundaries of intelligence," Alibaba CEO Eddie Wu said during the call. "Why is that the primary aim? Well, it's because of all the visible AI application scenarios today that we see around content creation, search and so on and so forth have arisen precisely as a result of the ongoing extension of those boundaries, and we want to keep pushing out those boundaries to create more and more opportunities."

Report: Alibaba to Release Upgraded Qwen 3 AI Model in Late April | PYMNTS.com

Alibaba Group Holding reportedly plans to release an upgraded version of its flagship AI model, Qwen 3, later this month. The timing of the release could change, Bloomberg reported Tuesday (April 1), citing an unnamed source. Alibaba did not immediately reply to PYMNTS' request for comment. The company's latest release will come at a time when the competition in the field is heating up, according to the Bloomberg report. Chinese companies have released several low-cost artificial intelligence (AI) services since DeepSeek gained attention with the release of its model that it said was less expensive to develop than those of its American rivals, the report said. OpenAI, Google and Anthropic have also released new AI models in recent weeks, per the report. Alibaba itself has already released some AI products this year, including a new model in its Qwen 2.5 series last week and a new version of its AI assistant Quark last month, according to the report. The company said in February that it will spend more on AI in the next three years than it has in the last decade. Alibaba management said during a Feb. 20 earnings call that the company's AI investments are based on the primary goal of achieving artificial general intelligence (AGI). "We aim to continue to develop models that extend the boundaries of intelligence," Eddie Wu, the company's CEO, said during the earnings call. "Why is that the primary aim? Well, it's because all of the visible AI application scenarios today that we see around content creation, search and so on and so forth have arisen precisely as a result of the ongoing extension of those boundaries, and we want to keep pushing out those boundaries to create more and more opportunities. When Alibaba Cloud announced the launch of the new Qwen2.5-Omni-7B on Thursday (March 27), it said this multimodal AI model can process inputs in the form of text, images, audio and video; can generate real-time responses in the form of text and natural speech; and can be deployed on mobile phones and laptops. Because the model is both compact and multimodal, it can power "agile, cost-effective AI agents," the company said.

Meet Qwen 2.5: Alibaba Unveils Powerful New AI Model to Rival Google and Meta

In the age of AI progression, all the companies are competing for the leading position in the AI race. While the LLM market is mostly dominated by Google's Gemini and Meta's language models, Alibaba has recently unveiled its new language model in the Qwen 2.5 family. Reportedly, the latest AI model that Alibaba has launched is about to mark a significant advancement in the company's AI capabilities and position it as a potent contender to the industry giants like Meta and Google. Alibaba has long been a potent contender in the . Previously as well, this company has claimed its Qwen 2.5 language models are a potent contender to the established forces of this industry. On Monday, Alibaba released , which has introduced multiple improvements over its predecessors, including enhanced coding capabilities and multilingual skills. Apart from these, it will also feature extended context processing. These additions will likely allow this latest model to perform complex tasks with more efficiency and fewer errors. As per the blog post by Alibaba, this is a vision-language model that has 32 billion parameters, and it demonstrates a massive improvement in image understanding. Additionally, it has a better reasoning-focused analysis skill that outperforms models from rivals. According to the official statement from Alibaba, "This unique combination makes it the perfect foundation for developing agile, cost-effective AI agents that deliver tangible value, especially intelligent voice applications."

Alibaba unveils new flagship AI model: Qwen2.5-Omni By Investing.com

Investing.com -- Alibaba Group Holdings Ltd ADR (NYSE:BABA) has introduced Qwen2.5-Omni, its new flagship model in the Qwen series. The end-to-end multimodal model is designed for extensive multimodal perception and can process a variety of inputs such as text, images, audio, and video. It provides real-time streaming responses through text generation and natural speech synthesis. Key features of the model include its Thinker-Talker architecture, designed to perceive a range of modalities, including text, images, audio, and video. This architecture allows the model to generate text and natural speech responses simultaneously. It also includes a novel position embedding, dubbed TMRoPE (Time-aligned Multimodal RoPE), which synchronizes the timestamps of video inputs with audio. The model is designed for fully real-time interactions, supporting chunked input and immediate output. It surpasses many existing streaming and non-streaming alternatives in terms of robustness and naturalness in speech generation. Qwen2.5-Omni showcases exceptional performance across all modalities and outperforms the similarly sized Qwen2-Audio in audio capabilities. It also matches the performance of Qwen2.5-VL-7B. Qwen2.5-Omni employs the Thinker-Talker architecture, where the Thinker functions like a brain, processing and understanding inputs from text, audio, and video modalities. It generates high-level representations and corresponding text. The Talker operates like a human mouth, taking in the high-level representations and text produced by the Thinker and outputting discrete tokens of speech fluidly. A comprehensive evaluation of Qwen2.5-Omni has been conducted, showing strong performance across all modalities when compared to similarly sized single-modality models and closed-source models like Qwen2.5-VL-7B, Qwen2-Audio, and Gemini-1.5-pro. In tasks requiring the integration of multiple modalities, such as OmniBench, Qwen2.5-Omni achieves state-of-the-art performance. In the near future, Alibaba plans to enhance the model's ability to follow voice commands and improve audio-visual collaborative understanding. The company also aims to integrate more modalities towards an omni-model. The Qwen2.5-Omni model is now publicly available on platforms like Hugging Face, ModelScope, DashScope, and GitHub. Users can experience the model's interactive features through a demo or join discussions on Discord.

Alibaba set to introduce upgraded AI model, Qwen 3, amid rising competition - Bloomberg By Investing.com

Investing.com -- Alibaba (NYSE:BABA) Group Holding Ltd., the ecommerce and cloud computing leader in China, is preparing to launch Qwen 3, an improved version of its primary AI model. The new offering is expected to be available later this month, although the exact date remains uncertain, according to a report from Bloomberg, citing a source close to the matter. Alibaba has been rapidly releasing AI products since deciding to fully embrace the technology earlier this year. Just last week, the company introduced a new model in its Qwen 2.5 series, capable of processing text, pictures, audio, and video. This model is efficient enough to operate directly on mobile phones and laptops. In addition, a new version of the AI assistant Quark app was launched last month by the company. The move comes as competition in the AI space intensifies, with rivals such as OpenAI and DeepSeek also releasing a stream of new models. DeepSeek, a fellow Hangzhou-based company, recently took the spotlight with an effective model that reportedly cost just a few million dollars to construct. This has spurred a surge of low-cost AI services from China's tech leaders, posing a potential challenge to premium US offerings from companies like OpenAI, Alphabet (NASDAQ:GOOGL) Inc.'s Google, and Microsoft Corp (NASDAQ:MSFT).