Alibaba Unveils Qwen 3: A New Family of Hybrid AI Reasoning Models Challenging Global Leaders

Alibaba unveils Qwen 3, a family of 'hybrid' AI reasoning models | TechCrunch

Chinese tech company Alibaba on Monday released Qwen 3, a family of AI models the company claims matches and in some cases outperforms the best models available from Google and OpenAI. Most of the models are -- or soon will be -- available for download under an "open" license from AI dev platform Hugging Face and GitHub. They range in size from 0.6 billion parameters to 235 billion parameters. Parameters roughly correspond to a model's problem-solving skills, and models with more parameters generally perform better than those with fewer parameters. The rise of China-originated model series like Qwen have increased the pressure on American labs such as OpenAI to deliver more capable AI technologies. They've also led policymakers to implement restrictions aimed at limiting the ability of Chinese AI companies to obtain the chips necessary to train models. According to Alibaba, Qwen 3 models are "hybrid" models in the sense that they can take time and "reason" through complex problems or answer simpler requests quickly. Reasoning enables the models to effectively fact-check themselves, similar to models like OpenAI's o3, but at the cost of higher latency. "We have seamlessly integrated thinking and non-thinking modes, offering users the flexibility to control the thinking budget," wrote the Qwen team in a blog post. The Qwen 3 models support 119 languages, Alibaba says, and were trained on a data set of nearly 36 trillion tokens. Tokens are the raw bits of data that the model processes; 1 million tokens is equivalent to about 750,000 words. Alibaba says Qwen 3 was trained on a combination of textbooks, "question-answer pairs," code snippets, and more. These improvements, along with others, greatly boosted Qwen 3's performance compared to its predecessor, Qwen 2, says Alibaba. On Codeforces, a platform for programming contests, the largest Qwen 3 model -- Qwen-3-235B-A22B -- beats out OpenAI's o3-mini. Qwen-3-235B-A22B also bests o3-mini on the latest version of AIME, a challenging math benchmark, and BFCL, a test for assessing a model's ability to "reason" about problems. But Qwen-3-235B-A22B isn't publicly available -- at least not yet. The largest public Qwen 3 model, Qwen3-32B, is still competitive with a number of proprietary and open AI models, including Chinese AI lab DeepSeek's R1. Qwen3-32B surpasses OpenAI's o1 model on several tests, including an accuracy benchmark called LiveBench. Alibaba says Qwen 3 "excels" in tool-calling capabilities as well as following instructions and copying specific data formats. In addition to releasing models for download, Qwen 3 is available from cloud providers including Fireworks AI and Hyperbolic. Tuhin Srivastava, co-founder and CEO of AI cloud host Baseten, said that Qwen 3 is another point in the trend line of open models keeping pace with closed-source systems such as OpenAI's. "The U.S. is doubling down on restricting sales of chips to China and purchases from China, but models like Qwen 3 that are state-of-the-art and open [...] will undoubtedly be used domestically," he told TechCrunch in a statement. "It reflects the reality that businesses are both building their own tools [as well as] buying off the shelf via closed-model companies like Anthropic and OpenAI."

Alibaba unveils advanced Qwen 3 AI as Chinese tech rivalry intensifies

April 29 (Reuters) - Chinese tech giant Alibaba Group (9988.HK), opens new tab launched Qwen 3 on Tuesday, an upgraded version of its flagship artificial intelligence model that introduces new hybrid reasoning capabilities. The launch comes as competition in China's AI sector intensifies, spurred by the breakout success of local startup DeepSeek earlier this year, which claimed to have built high-performing models at lower costs than their Western counterparts. Chinese search leader Baidu (9888.HK), opens new tab joined the AI arms race last Friday with the release of its Ernie 4.5 Turbo and reasoning-focused Ernie X1 Turbo models. Alibaba's newest release merges conventional AI functions with advanced dynamic reasoning, creating what the company calls a more adaptable and efficient platform for app and software developers. The e-commerce giant had previously rushed out its Qwen 2.5-Max model in late January, just days after DeepSeek's announcement, claiming superior performance. Reporting by Jasmeen Ara Shaikh in Bengaluru; Editing by Sherry Jacob-Phillips Our Standards: The Thomson Reuters Trust Principles., opens new tab Suggested Topics:Artificial Intelligence

Alibaba launches new Qwen LLMs in China's latest open-source AI breakthrough

Qwen3 is Alibaba's debut into so-called "hybrid reasoning models," which it says combines traditional LLM capabilities with "advanced, dynamic reasoning." Alibaba released the next generation of its open-sourced large language models, Qwen3, on Tuesday -- and experts are calling it yet another breakthrough in China's booming open-source artificial intelligence space. In a blog post, the Chinese tech giant said Qwen3 promises improvements in reasoning, instruction following, tool usage and multilingual tasks, rivaling other top-tier models such as DeepSeek's R1 in several industry benchmarks. The LLM series includes eight variations that span a range of architectures and sizes, offering developers flexibility when using Qwen to build AI applications for edge devices like mobile phones. Qwen3 is also Alibaba's debut into so-called "hybrid reasoning models," which it says combines traditional LLM capabilities with "advanced, dynamic reasoning." According to Alibaba, such models can seamlessly transition between a "thinking mode" for complex tasks such as coding and a "non-thinking mode" for faster, general-purpose responses. "Notably, the Qwen3-235B-A22B MoE model significantly lowers deployment costs compared to other state-of-the-art models, reinforcing Alibaba's commitment to accessible, high-performance AI," Alibaba said. The new models are already freely available for individual users on platforms like Hugging Face and GitHub, as well as Alibaba Cloud's web interface. Qwen3 is also being used to power Alibaba's AI assistant, Quark.

Alibaba is launching its own AI reasoning models to compete with DeepSeek

Yet another player is claiming the top spot for AI. Alibaba, one of the world's largest e-commerce companies, has announced the launch of its latest AI reasoning model, known as Qwen 3. Alibaba claims these models match (and even outperform) models from both Google and OpenAI in reasoning tasks. Qwen itself isn't new (as the 3 in the name suggests), but this is the first time Alibaba has really tried to challenge the top dogs, as well as taking its first steps into more complex reasoning tasks. Most of the models will soon be available for download under an open license from AI dev companies Hugging Face or GitHub. There will be a few models ranging in their parameters, going from 0.6 billion to 235 billion. Parameters roughly translate to a model's ability to problem solve in this instance. The more parameters, the better its ability, but also the longer it takes and the more it costs. While they haven't announced any further details, it is likely these models will eventually make their way over to the Qwen chatbot. The models offer two modes. In thinking mode, the model will take time, going step by step to respond with the best answer. In non-thinking mode, the model provides quick responses to simpler questions where speed is more important than depth. "This flexibility allows users to control how much 'thinking' the model performs based on the task at hand. For example, harder problems can be tackled with extended reasoning, while easier ones can be answered directly without delay," said the Qwen team in a blog post. "Crucially, the integration of these two modes greatly enhances the model's ability to implement stable and efficient thinking budget control." Alongside DeepSeek, Alibaba is a sign that China is catching up with American AI companies. DeepSeek has already performed at the same level, if not better, than the likes of ChatGPT and Google on reasoning tasks. If Alibaba's claims are true, this is yet another set of reasoning models that can match ChatGPT's performance. Interestingly, it would also be a more powerful, smarter model than what DeepSeek has created. Like DeepSeek, Alibaba seems to undercut the US market, offering AI technology that is just as powerful but cheaper. This does, however, raise questions of ethics and security as several AI companies have been accused of cutting corners. AI companies are moving to a slightly new system for AI models. Originally, there was just one version of an AI system, now there are multiple. These different systems are built to handle different tasks, putting in more effort for complicated tasks and using less energy for easy challenges. ChatGPT, for example, has ChatGPT 4o, which is its general model for most tasks, but also 4o mini, for everyday simple requests. On top of that, it has a range of reasoning models. Reasoning models are designed to take on complicated tasks. This is everything from coding to multi-step requests (asking it to do several things in a row) and dealing with deep research projects. With this latest update from Qwen, Alibaba is aiming to do the same, offering a model that can do the simple everyday tasks, but also achieve the complicated thinking that is becoming expected of AI today.

Alibaba launches open source Qwen3 model that surpasses OpenAI o1 and DeepSeek R1

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Chinese e-commerce and web giant Alibaba's Qwen team has officially launched a new series of open source AI large language multimodal models known as Qwen3 that appear to be among the state-of-the-art for open models, and approach performance of proprietary models from the likes of OpenAI and Google. The Qwen3 series features two "mixture-of-experts" models and six dense models for a total of eight (!) new models. The "mixture-of-experts" approach involves having several different specialty model types combined into one, with only those relevant models to the task at hand being activated when needed in the internal settings of the model (known as parameters). It was popularized by open source French AI startup Mistral. According to the team, the 235-billion parameter version of Qwen3 codenamed A22B outperforms DeepSeek's open source R1 and OpenAI's proprietary o1 on key third-party benchmarks including ArenaHard (with 500 user questions in software engineering and math) and nears the performance of the new, proprietary Google Gemini 2.5-Pro. Overall, the benchmark data positions Qwen3-235B-A22B as one of the most powerful publicly available models, achieving parity or superiority relative to major industry offerings. Hybrid (reasoning) theory The Qwen3 models are trained to provide so-called "hybrid reasoning" or "dynamic reasoning" capabilities, allowing users to toggle between fast, accurate responses and more time-consuming and compute-intensive reasoning steps (similar to OpenAI's "o" series) for more difficult queries in science, math, engineering and other specialized fields. This is an approach pioneered by Nous Research and other AI startups and research collectives. With Qwen3, users can engage the more intensive "Thinking Mode" using the button marked as such on the Qwen Chat website or by embedding specific prompts like or when deploying the model locally or through the API, allowing for flexible use depending on the task complexity. Users can now access and deploy these models across platforms like Hugging Face, ModelScope, Kaggle, and GitHub, as well as interact with them directly via the Qwen Chat web interface and mobile applications. The release includes both Mixture of Experts (MoE) and dense models, all available under the Apache 2.0 open-source license. In my brief usage of the Qwen Chat website so far, it was able to generate imagery relatively rapidly and with decent prompt adherence -- especially when incorporating text into the image natively while matching the style. However, it often prompted me to log in and was subject to the usual Chinese content restrictions (such as prohibiting prompts or responses related to the Tiananmen Square protests). In addition to the MoE offerings, Qwen3 includes dense models at different scales: Qwen3-32B, Qwen3-14B, Qwen3-8B, Qwen3-4B, Qwen3-1.7B, and Qwen3-0.6B. These models vary in size and architecture, offering users options to fit diverse needs and computational budgets. The Qwen3 models also significantly expand multilingual support, now covering 119 languages and dialects across major language families. This broadens the models' potential applications globally, facilitating research and deployment in a wide range of linguistic contexts. Model training and architecture In terms of model training, Qwen3 represents a substantial step up from its predecessor, Qwen2.5. The pretraining dataset doubled in size to approximately 36 trillion tokens. The data sources include web crawls, PDF-like document extractions, and synthetic content generated using previous Qwen models focused on math and coding. The training pipeline consisted of a three-stage pretraining process followed by a four-stage post-training refinement to enable the hybrid thinking and non-thinking capabilities. The training improvements allow the dense base models of Qwen3 to match or exceed the performance of much larger Qwen2.5 models. Deployment options are versatile. Users can integrate Qwen3 models using frameworks such as SGLang and vLLM, both of which offer OpenAI-compatible endpoints. For local usage, options like Ollama, LMStudio, MLX, llama.cpp, and KTransformers are recommended. Additionally, users interested in the models' agentic capabilities are encouraged to explore the Qwen-Agent toolkit, which simplifies tool-calling operations. Junyang Lin, a member of the Qwen team, commented on X that building Qwen3 involved addressing critical but less glamorous technical challenges such as scaling reinforcement learning stably, balancing multi-domain data, and expanding multilingual performance without quality sacrifice. Lin also indicated that the team is transitioning focus toward training agents capable of long-horizon reasoning for real-world tasks. What it means for enterprise decision-makers Engineering teams can point existing OpenAI-compatible endpoints to the new model in hours instead of weeks. The MoE checkpoints (235 B parameters with 22 B active, and 30 B with 3 B active) deliver GPT-4-class reasoning at roughly the GPU memory cost of a 20-30 B dense model. Official LoRA and QLoRA hooks allow private fine-tuning without sending proprietary data to a third-party vendor. Dense variants from 0.6 B to 32 B make it easy to prototype on laptops and scale to multi-GPU clusters without rewriting prompts. Running the weights on-premises means all prompts and outputs can be logged and inspected. MoE sparsity reduces the number of active parameters per call, cutting the inference attack surface. The Apache-2.0 license removes usage-based legal hurdles, though organizations should still review export-control and governance implications of using a model trained by a China-based vendor. Yet at the same time, it also offers a viable alternative to other Chinese players including DeepSeek, Tencent, and ByteDance -- as well as the myriad and growing number of North American models such as the aforementioned OpenAI, Google, Microsoft, Anthropic, Amazon, Meta and others. The permissive Apache 2.0 license -- which allows for unlimited commercial usage -- is also a big advantage over other open source players like Meta, whose licenses are more restrictive. It indicates furthermore that the race between AI providers to offer ever-more powerful and accessible models continues to remain highly competitive, and savvy organizations looking to cut costs should attempt to remain flexible and open to evaluating said new models for their AI agents and workflows. Looking ahead The Qwen team positions Qwen3 not just as an incremental improvement but as a significant step toward future goals in Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI), AI significantly smarter than humans. Plans for Qwen's next phase include scaling data and model size further, extending context lengths, broadening modality support, and enhancing reinforcement learning with environmental feedback mechanisms. As the landscape of large-scale AI research continues to evolve, Qwen3's open-weight release under an accessible license marks another important milestone, lowering barriers for researchers, developers, and organizations aiming to innovate with state-of-the-art LLMs.

Qwen swings for a double with 2.5-Omni-3B model that runs on consumer PCs, laptops

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Chinese e-commerce and cloud giant Alibaba isn't taking the pressure off other AI model providers in the U.S. and abroad. Just days after releasing its new, state-of-the-art open source Qwen3 large reasoning model family, Alibaba's Qwen team today released Qwen2.5-Omni-3B, a lightweight version of its preceding multimodal model architecture designed to run on consumer-grade hardware without sacrificing broad functionality across text, audio, image, and video inputs. Qwen2.5-Omni-3B is a scaled-down, 3-billion-parameter variant of the team's flagship 7 billion parameter (7B) model. (Recall parameters refer to the number of settings governing the model's behavior and functionality, with more typically denoting more powerful and complex models). While smaller in size, the 3B version retains over 90% of the larger model's multimodal performance and delivers real-time generation in both text and natural-sounding speech. A major improvement comes in GPU memory efficiency. The team reports that Qwen2.5-Omni-3B reduces VRAM usage by over 50% when processing long-context inputs of 25,000 tokens. With optimized settings, memory consumption drops from 60.2 GB (7B model) to just 28.2 GB (3B model), enabling deployment on 24GB GPUs commonly found in high-end desktops and laptop computers -- instead of the larger dedicated GPU clusters or workstations found in enterprises. According to the developers, it achieves this through architectural features such as the Thinker-Talker design and a custom position embedding method, TMRoPE, which aligns video and audio inputs for synchronized comprehension. However, the licensing terms specify for research only -- meaning enterprises cannot use the model to build commercial products unless they obtain a separate license from Alibaba's Qwen Team, first. The announcement follows increasing demand for more deployable multimodal models and is accompanied by performance benchmarks showing competitive results relative to larger models in the same series. The model is now freely available for download from: Developers can integrate the model into their pipelines using Hugging Face Transformers, Docker containers, or Alibaba's vLLM implementation. Optional optimizations such as FlashAttention 2 and BF16 precision are supported for enhanced speed and reduced memory consumption. Benchmark performance shows strong results even approaching much larger parameter models Despite its reduced size, Qwen2.5-Omni-3B performs competitively across key benchmarks: The narrow performance gap in video and speech tasks highlights the efficiency of the 3B model's design, particularly in areas where real-time interaction and output quality matter most. Real-time speech, voice customization, and more Qwen2.5-Omni-3B supports simultaneous input across modalities and can generate both text and audio responses in real time. The model includes voice customization features, allowing users to choose between two built-in voices -- Chelsie (female) and Ethan (male) -- to suit different applications or audiences. Users can configure whether to return audio or text-only responses, and memory usage can be further reduced by disabling audio generation when not needed. Community and ecosystem growth The Qwen team emphasizes the open-source nature of its work, providing toolkits, pretrained checkpoints, API access, and deployment guides to help developers get started quickly. The release also follows recent momentum for the Qwen2.5-Omni series, which has reached top rankings on Hugging Face's trending model list. Junyang Lin from the Qwen team commented on the motivation behind the release on X, stating, "While a lot of users hope for smaller Omni model for deployment we then build this." What it means for enterprise technical decision-makers For enterprise decision makers responsible for AI development, orchestration, and infrastructure strategy, the release of Qwen2.5-Omni-3B may appear, at first glance, like a practical leap forward. A compact, multimodal model that performs competitively against its 7B sibling while running on 24GB consumer GPUs offers real promise in terms of operational feasibility. But as with any open-source technology, licensing matters -- and in this case, the license draws a firm boundary between exploration and deployment. The Qwen2.5-Omni-3B model is licensed for non-commercial use only under Alibaba Cloud's Qwen Research License Agreement. That means organizations can evaluate the model, benchmark it, or fine-tune it for internal research purposes -- but cannot deploy it in commercial settings, such as customer-facing applications or monetized services, without first securing a separate commercial license from Alibaba Cloud. For professionals overseeing AI model lifecycles -- whether deploying across customer environments, orchestrating at scale, or integrating multimodal tools into existing pipelines -- this restriction introduces important considerations. It may shift Qwen2.5-Omni-3B's role from a deployment-ready solution to a testbed for feasibility, a way to prototype or evaluate multimodal interactions before deciding whether to license commercially or pursue an alternative. Those in orchestration and ops roles may still find value in piloting the model for internal use cases -- like refining pipelines, building tooling, or preparing benchmarks -- so long as it remains within research bounds. Data engineers or security leaders might likewise explore the model for internal validation or QA tasks, but should tread carefully when considering its use with proprietary or customer data in production environments. The real takeaway here may be about access and constraint: Qwen2.5-Omni-3B lowers the technical and hardware barrier to experimenting with multimodal AI, but its current license enforces a commercial boundary. In doing so, it offers enterprise teams a high-performance model for testing ideas, evaluating architectures, or informing make-vs-buy decisions -- yet reserves production use for those willing to engage Alibaba for a licensing discussion. In this context, Qwen2.5-Omni-3B becomes less a plug-and-play deployment option and more a strategic evaluation tool -- a way to get closer to multimodal AI with fewer resources, but not yet a turnkey solution for production.

Alibaba claims leadership in AI reasoning with latest Qwen3 models - SiliconANGLE

Alibaba claims leadership in AI reasoning with latest Qwen3 models Chinese technology giant Alibaba Group Holding Ltd. has announced Qwen3, a new family of artificial intelligence models that it says can outperform competing models from companies such as OpenAI and Google LLC. The new release underscores the rapid pace of development within China's AI industry since DeepSeek Ltd. first burst onto the scene late last year. The e-commerce giant said the new Qwen3 models surpass the capabilities of DeepSeek's best models in a number of areas, including coding and math-based problems. It's releasing a number of models within the Qwen3 family under an open-source license, ranging in size from 600 million to 235 billion parameters, which is a measure that roughly corresponds to problem-solving abilities. As a rule, the more parameters a model has, the better it performs. Within the new Qwen3 series are two "mixture-of-experts" or MoE models that Alibaba says are able to compete with the most advanced reasoning models launched by Google and Anthropic PBC. Reasoning models are designed to mimic the way humans think about problems, taking more time to consider things and perform fact checking for accuracy. By using the MoE technique, AI models can enhance their reasoning skills by dividing a task into smaller segments, similar to how a company might employ teams of specialists to focus on specific parts of a more challenging problem. By splitting the tasks across different parts of the model, the problem solving process becomes more efficient. "We have seamlessly integrated thinking and non-thinking modes, offering users the flexibility to control the thinking budget," Alibaba's Qwen team said in a blog post. "This design enables users to configure task-specific budgets with greater ease." Alibaba said the Qwen3 models support 119 languages and have been trained on a dataset containing almost 36 trillion tokens, which are the raw bits of data they process while being "taught". One million tokens is equivalent to around 750,000 words, and in this case the data was drawn from various textbooks, code snippets, AI-generated data, question-answer pairs and other sources. In various benchmark tests, Alibaba's Qwen3 models delivered some impressive results, edging out recent "high-end" models from U.S.-based AI companies, such as OpenAI's o3-mini and o4-mini models. For instance, on the Codeforces benchmark that measures models' ability to write code, the largest Qwen-3-235B-A22B model surpassed o3-mini and Google's Gemini 2.5 Pro. It also beat 03-mini on the AIME mathematics benchmark, as well as the BFCL test that assesses AI models' reasoning abilities. Since DeepSeek's R1 reasoning model first burst onto the scene at the end of December, upstaging OpenAI despite being developed at just a fraction of the cost, Chinese tech leaders have released a flurry of similarly powerful AI models. Alibaba launched the Qwen-2.5 series models just a few weeks earlier, noting they can process multimodal data formats including text, images, audio and video. Those models are notably lightweight, designed to run directly on smartphones and laptops. The emergence of powerful, open-source Chinese AI models has upped the ante for U.S. AI companies, which were until recently seen as industry leaders. But their status is coming under threat, especially because American-made models are generally trained at much higher costs than their Chinese counterparts. The U.S. government has also responded, introducing further sanctions that aim to prevent Chinese companies from getting their hands on the powerful graphics processing units used to train and run AI models. In its most recent move, the U.S. slapped an export license on Nvidia Corp.'s H20 GPU, which had been designed specifically to comply with earlier sanctions on China. OpenAI has responded by saying it will release an "open-weights" reasoning model in the next few months, marking a dramatic reversal from its usual approach, where the inner workings of its models are essentially a "black box". Alibaba Chief Executive Officer Eddie Wu said in February that the company's main objective is to build an "artificial general intelligence" system that will ultimately surpass the intellectual capabilities of humans.

Alibaba's Qwen3 Outperforms OpenAI's o1 and o3-mini, on Par With Gemini 2.5 Pro | AIM

Chinese giant Alibaba has released the Qwen3 family of open-weight AI models. Apart from its flagship 235B (billion) parameter model, Qwen3 is released in various sizes. It includes variants consisting of 0.6B, 1.7B, 4B, 8B, 14B, 32B, parameters, alongside a 235B model with 22B activated parameters, and a 30B model with 3 billion activated parameters. The model can be deployed locally using tools such as Ollama and LM Studio, and can also be accessed through a web browser using Qwen Chat. Users can switch between a "thinking" mode for tasks that require reasoning and a "non-thinking" mode for tasks that demand quick responses. Qwen3's 235B parameter model outperforms OpenAI's o1 and o3-mini (medium) reasoning models on benchmarks that evaluate its abilities in mathematical and programming tasks. Besides, it also offers performance parity with Google's Gemini 2.5 Pro models on several benchmarks. Having said that, the model lags behind OpenAI's newly released o4-mini (high) model. In the LiveCodeBench coding benchmark, Qwen3's 235B scores 70.7%, whereas the o4-mini (high) scored 80%. On the AIME 2024 math benchmark, OpenAI's o4-mini (high) scored 94% -- only slightly higher than Qwen3's 235B (85.7%). Benchmark scores of other newly released models can be found on Artificial Analysis. Moreover, other variants of the Qwen-3 models outperformed their predecessors, and the 30B parameter variant outperformed DeepSeek-V3 and OpenAI's GPT-4o on benchmarks. Simon Willison, co-creator of the Django Web Framework, said in a blog post, "The thing that stands out most to me about the Qwen3 release is how well coordinated it was across the LLM ecosystem." While Wilson was using the models, he noted that they "worked directly" with all popular LLM serving frameworks from the day of their release. "This is an extraordinary level of coordination for a model release! I haven't seen any other model providers make this level of effort -- the usual pattern is to dump a bunch of models on Hugging Face for a single architecture (usually NVIDIA) and then wait for the community to catch up with quantisations and conversions for everything else," he added. Besides, given the spectrum of sizes in which the model has been released, Wilson said, "0.6B and 1.7B should run fine on an iPhone, and 32B will fit on my 64GB Mac with room to spare for other applications." The Qwen3 family of models is a successor to the Qwen2.5 models. Last month, the company announced the QwQ 32 billion parameter model, which was said to achieve comparable performance with DeepSeek-R1, despite being a much smaller model. Furthermore, the company also launched the QwQ-Max-Preview model last month, built on the Qwen2.5 Max. The model was said to specialise in mathematics and coding-based tasks.

Alibaba launches its own DeepSeek, ChatGPT rival Qwen3

As the AI race heats up in China, Alibaba becomes the latest to claim a DeepSeek-rivalling LLM model There's little question that the release of China's DeepSeek R1 in January of this year significantly accelerated the global race to grab a part of this burgeoning AI marketplace. Now China's Alibaba has signalled its intention to be a serious player, with the release of Qwen3, the latest generation of its open-sourced large language model (LLM) family, which it claims will rival the best in the market. Alibaba says Qwen3 was trained on a huge dataset of 36 trillion tokens, double that of its predecessor Qwen2.5. "Qwen3 delivers significant advancement on reasoning, instruction following, tool use and multilingual tasks," Alibaba claims. It has made Qwen3 models available for download on Hugging Face, GitHub and Modelscope, as well as being available on its own site chat.qwen.ai. Alibaba plans to make API access available soon through its AI model development platform Model Studio. Qwen3 also powers Alibaba's "AI super assistant" application, Quark. According to Alibaba, the Qwen model family has attracted over 300 million downloads worldwide since its debut, with developers creating more than 100,000 Qwen-based derivative models on Hugging Face. Now its hoping this new release will rival the best models today. Just last week, the latest AI Index 2025 Annual Report by the Institute for Human-Centered AI at Stanford University found that, while the US is still the leader when it comes to producing top AI models - US-based institutions produced 40 notable AI models in 2024, it said, significantly outpacing China's 15 and Europe's three - other countries are catching up when it comes to quality. According to the AI Index, Google and OpenAI remained the leaders in the sector, but it cautioned they were facing tougher competition not alone from US companies like Meta and Anthropic, but of course from China's DeepSeek, which took the tech world by storm with the quality of its DeepSeek R1 model released in January. Now Chinese players like Butterfly Effect (Manus) and Alibaba are hoping to be a part of the conversation too. Don't miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic's digest of need-to-know sci-tech news.

Qwen3 shows open models can still rival closed giants

Alibaba released Qwen3, a family of AI models that the company claims matches and sometimes outperforms Google and OpenAI's best models, on Monday. The models range in size from 0.6 billion parameters to 235 billion parameters and are available for download under an "open" license from AI dev platform Hugging Face and GitHub. The Qwen3 models are described as "hybrid" because they can take time to "reason" through complex problems or answer simpler requests quickly. This reasoning ability enables the models to fact-check themselves effectively, similar to OpenAI's o3 model, but with higher latency. According to the Qwen team, they have "seamlessly integrated thinking and non-thinking modes, offering users the flexibility to control the thinking budget." Some Qwen3 models adopt a mixture of experts (MoE) architecture, which can be more computationally efficient for answering queries. MoE breaks down tasks into subtasks and delegates them to smaller, specialized "expert" models. The models support 119 languages and were trained on a dataset of nearly 36 trillion tokens, including textbooks, question-answer pairs, code snippets, and AI-generated data. Alibaba claims that Qwen3's capabilities have greatly improved compared to its predecessor, Qwen2. The largest Qwen3 model, Qwen-3-235B-A22B, performs competitively on benchmark evaluations, beating OpenAI's o3-mini and Google's Gemini 2.5 Pro on Codeforces, a platform for programming contests. It also outperforms o3-mini on the latest version of AIME, a challenging math benchmark, and BFCL, a test for assessing a model's ability to reason about problems. While Qwen-3-235B-A22B is not publicly available, the largest public Qwen3 model, Qwen3-32B, is still competitive with several proprietary and open AI models. Qwen3-32B surpasses OpenAI's o1 model on several tests, including the coding benchmark LiveCodeBench. Alibaba says Qwen3 excels in tool-calling capabilities, following instructions, and copying specific data formats. Tuhin Srivastava, co-founder and CEO of AI cloud host Baseten, noted that Qwen3 is another example of open models keeping pace with closed-source systems like OpenAI's. He added that models like Qwen3 will likely be used domestically, despite U.S. restrictions on chip sales to China. Qwen3 is available for download and through cloud providers, including Fireworks AI and Hyperbolic.

Alibaba unveils advanced Qwen 3 AI as Chinese tech rivalry intensifies

Alibaba has launched Qwen 3, an upgraded AI model with hybrid reasoning abilities, amid rising competition in China's AI sector. The release follows moves by rivals like DeepSeek and Baidu. Qwen 3 aims to offer greater adaptability and efficiency for developers, building on Alibaba's earlier Qwen 2.5-Max model.Chinese tech giant Alibaba Group launched Qwen 3 on Tuesday, an upgraded version of its flagship artificial intelligence model that introduces new hybrid reasoning capabilities. The launch comes as competition in China's AI sector intensifies, spurred by the breakout success of local startup DeepSeek earlier this year, which claimed to have built high-performing models at lower costs than their Western counterparts. Chinese search leader Baidu joined the AI arms race last Friday with the release of its Ernie 4.5 Turbo and reasoning-focused Ernie X1 Turbo models. Alibaba's newest release merges conventional AI functions with advanced dynamic reasoning, creating what the company calls a more adaptable and efficient platform for app and software developers. The ecommerce giant had previously rushed out its Qwen 2.5-Max model in late January, just days after DeepSeek's announcement, claiming superior performance.

China's Alibaba Dunks on Meta with New Qwen 3 AI Models

The smaller, Qwen3-30B-A3B MoE model, with only 3 billion activated parameters, beats GPT-4o by a significant margin. The Chinese tech giant, Alibaba, has launched eight new open-weight AI models under the Qwen 3 series. New Qwen 3 models include two MoE (Mixture of Experts) models, such as Qwen3-235B-A22B and Qwen3-30B-A3B. Qwen3-235B-A22B is the largest and flagship model, with a total of 235 billion parameters, and 22 billion activated parameters. Qwen3-30B-A3B is a smaller MoE model with a total of 30 billion parameters and 3 billion activated parameters. Apart from that, there are six dense models under the Qwen 3 series including Qwen3-32B, Qwen3-14B, Qwen3-8B, Qwen3-4B, Qwen3-1.7B, and Qwen3-0.6B. All Qwen 3 models support Hybrid Thinking Modes, which means they are both reasoning AI models and traditional LLMs. In the Thinking mode, the model can reason step by step, and in the Non-Thinking mode, the model provides a quick response. In addition, Qwen 3 models support over 119 languages and dialects from all around the world. It's one of the most diverse multilingual models out there. Next, Alibaba has worked to improve MCP support for Qwen 3 models, unlocking agentic capabilities further. As for performance, the largest Qwen3-235B-A22B model delivers competitive results along the lines of DeepSeek R1, Grok 3 Beta, Gemini 2.5 Pro, and OpenAI o1. What I find interesting is that the smaller Qwen3-30B-A3B model, with only 3 billion activated parameters, outperforms DeepSeek V3 and OpenAI's GPT-4o model. Alibaba says Qwen 3 models offer great performance in coding, math, science, and general capabilities. Overall, Qwen 3 represents a family of highly capable, frontier AI models from China. Now with the upcoming DeepSeek R2, China is well-positioned to rival Western AI labs.

Qwen 3 Open Source Hybrid AI Beats Deepseek R1 : Performance Fully Tested

What if the future of artificial intelligence wasn't locked behind corporate walls but instead placed directly in your hands? Enter Qwen 3, Alibaba's latest open source hybrid large language model (LLM) that's not just a contender but a disruptor in the AI world. With a staggering 235 billion parameters at its peak, Qwen 3 doesn't just compete with proprietary giants like Deepseek R1 -- it outperforms them in critical benchmarks, from coding to logical reasoning. Imagine a model that activates only a fraction of its parameters during operation, delivering unmatched efficiency without compromising power. This isn't just another AI release; it's a bold redefinition of what open source innovation can achieve. In this coverage, World of AI explore how Qwen 3's new Mixture-of-Experts architecture and hybrid thinking mode are setting new standards for efficiency and adaptability. You'll discover why its open source Apache 2.0 license is a fantastic option for developers and organizations alike, offering unparalleled freedom to customize and deploy. From its multilingual support spanning 119 languages to its ability to solve complex mathematical problems with precision, Qwen 3 is designed to meet the demands of industries that value accuracy and scalability. Qwen 3's flagship model features an impressive 235 billion parameters, with only 22 billion active during any given operation. This efficiency is achieved through its MoE architecture, which optimizes computational resources while maintaining high performance. For users seeking lighter alternatives, the series includes a 30-billion-parameter model with just 3 billion active parameters. Additionally, six dense models, ranging from 0.6 to 32 billion parameters, offer flexibility to address a wide range of use cases. All models are distributed under the Apache 2.0 license, making sure unrestricted access and adaptability. This licensing framework enables developers and organizations to modify, deploy, and integrate Qwen 3 into their workflows without restrictive barriers. By fostering collaboration and innovation, the open source nature of Qwen 3 positions it as a valuable tool for advancing AI development. Qwen 3 has demonstrated exceptional performance across multiple benchmarks, surpassing competitors like Deepseek R1, Gro 3 Gemini 2.5 Pro, and OpenAI models. Its strengths are particularly evident in areas such as: While Qwen 3's creative capabilities, such as storytelling or artistic content generation, have shown mixed results, its logical reasoning and problem-solving skills remain unparalleled. This makes it an ideal choice for industries that prioritize precision, such as finance, engineering, and scientific research. Here is a selection of other guides from our extensive library of content you may find of interest on the Qwen Large Language Models (LLM). Qwen 3 introduces several advanced features that enhance its efficiency and adaptability, making it a standout in the AI landscape: These features make Qwen 3 an attractive option for both large-scale deployments and localized installations. By balancing performance with resource efficiency, it addresses the needs of organizations with varying computational capacities and operational requirements. Qwen 3 has been rigorously tested across diverse domains, showcasing its versatility and reliability. Key applications include: Although its performance in creative tasks like storytelling or artistic content generation is less consistent, Qwen 3 excels in logic-driven and structured challenges. This makes it a valuable asset for industries that demand accuracy and efficiency, such as technology, healthcare, and education. One of Qwen 3's most notable features is its scalability and ease of deployment. The model is optimized for rapid integration, allowing organizations to incorporate it into their systems with minimal overhead. Its open weights allow for local installation, granting users greater control over their AI solutions. This is particularly beneficial for enterprises with strict data privacy requirements or limited access to cloud infrastructure. By offering both large-scale and localized deployment options, Qwen 3 ensures that organizations of all sizes can use its capabilities. Its adaptability makes it suitable for a wide range of applications, from enterprise-level solutions to individual projects. Qwen 3 represents a significant advancement in the field of open source AI. By combining efficiency, scalability, and robust performance, it establishes a new benchmark for hybrid LLMs. Its innovative features, such as the MoE architecture and hybrid thinking mode, are likely to influence the design of future AI models. For developers, researchers, and organizations, Qwen 3 offers a powerful alternative to proprietary models. Its open source nature ensures that the development of AI remains collaborative and inclusive, fostering innovation across the industry. As the AI landscape continues to evolve, Qwen 3 stands out as a model that prioritizes efficiency, adaptability, and accessibility, paving the way for broader adoption and impact.