A technical innovation has allowed Alibaba Group Holding, one of the leading players in China’s artificial intelligence boom, to develop a new generation of foundation models that match the strong performance of larger predecessors while being significantly smaller and more cost efficient.
Alibaba Cloud, the AI and cloud computing division of Alibaba, unveiled on Friday a new generation of large language models that it said heralded “the future of efficient LLMs”. The new models are nearly 13 times smaller than the company’s largest AI model, released just a week earlier.
Despite its compact size, Qwen3-Next-80B-A3B is among Alibaba’s best models to date, according to developers. The key lies in its efficiency: the model is said to perform 10 times faster in some tasks than the preceding Qwen3-32B released in April, while achieving a 90 per cent reduction in training costs.
Do you have questions about the biggest topics and trends from around the world? Get the answers with SCMP Knowledge, our new platform of curated content with explainers, FAQs, analyses and infographics brought to you by our award-winning team.
Emad Mostaque, co-founder of the UK-based start-up Stability AI, said on X that Alibaba’s new model outperformed “pretty much any model from last year” despite an estimated training cost of less than US$500,000.
For comparison, training Google’s Gemini Ultra, released in February 2024, cost an estimated US$191 million, according to Stanford University’s AI Index.
Alibaba says its new generation of AI foundation models heralds the “the future of efficient LLMs”. Photo: Handout alt=Alibaba says its new generation of AI foundation models heralds the “the future of efficient LLMs”. Photo: Handout>
Artificial Analysis, a leading AI benchmarking firm, said Qwen3-Next-80B-A3B surpassed the latest versions of both DeepSeek R1 and Alibaba-backed start-up Moonshot AI’s Kimi-K2. Alibaba owns the South China Morning Post.
Several AI researchers attributed the success of Alibaba’s new model to a relatively new technique called “hybrid attention”.
Existing models face diminishing returns on efficiency as input lengths increase because of the way AI models determine which inputs are the most relevant. This “attention” mechanism involves trade-offs: better attention accuracy leads to higher computational expenses.
Those costs compound when models handle long context inputs, making it expensive to train sophisticated AI agents that autonomously execute tasks for users.
Qwen3-Next-80B-A3B addresses this challenge by incorporating a technique known as “Gated DeltaNet”, first proposed by researchers at the Massachusetts Institute of Technology and Nvidia in March.
Gated DeltaNet enhanced the model’s attention by making targeted adjustments to the input data and determining what information to retain and what to discard, said Zhou Peilin, an AI researcher at the Hong Kong University of Science and Technology.
This results in an accurate yet cost-effective attention mechanism. Citing scores from the Ruler benchmark, which evaluates AI models based on their ability to manage varying input lengths, Alibaba said Qwen3-Next-80B-A3B was comparable to its most powerful model, the Qwen3-235B-A22B-Thinking-2507, despite being smaller and more affordable.
Alibaba uses a technique known as “Gated DeltaNet” to develop its latest AI models. Photo: Handout alt=Alibaba uses a technique known as “Gated DeltaNet” to develop its latest AI models. Photo: Handout>
“It’s great to see that our DeltaNets … have been greatly scaled up by Alibaba to build excellent AI models,” said Juergen Schmidhuber, computer science professor at the King Abdullah University of Science and Technology, who contributed to the development of DeltaNets in the 1990s.
Qwen3-Next-80B-A3B also uses the “mixture-of-experts” (MoE) architecture, which has driven many efficiency gains in Chinese AI models over the past year, including DeepSeek-V3 and Moonshot’s Kimi-K2.
The MoE architecture divides a model into separate sub-networks or “experts” that specialise in subsets of input data to collaboratively perform tasks.
Alibaba enhanced the “sparsity” of its latest MoE architecture to improve efficiency. While DeepSeek-V3 and Kimi-K2 employ 256 and 384 experts, respectively, Qwen3-Next-80B-A3B features 512 experts but activates only 10 at a time.
According to Artificial Analysis, those innovations helped the model achieve parity with DeepSeek-V3.1, despite having just 3 billion active parameters compared with the latter’s 37 billion. Generally, a higher number of parameters indicates a more powerful model, but it also increases training and operational costs.
The efficiency gains are evident on Alibaba’s cloud platform, where the new model costs less to run than the Qwen3-235B-2507, which contains 235 billion parameters.
The new model architecture reflects a growing interest in smaller but more efficient AI models, amid rising concerns about the costs associated with scaling up the industry’s largest models.
According to AI research firm Epoch AI, the most expensive training run to date was xAI’s Grok 4, which cost US$490 million, with future training runs expected to exceed US$1 billion by 2027.
In August, researchers at Nvidia published a paper advocating for small language models as the future of agentic AI because of their flexibility and efficiency. The company is also experimenting with the Gated DeltaNet technique on its Nemotron models.
Meanwhile, Chinese AI giants are pushing for broader adoption of their models by ensuring they are small enough to run on laptops and smartphones.
Last month, Tencent Holdings launched four open-source AI models, each under 7 billion parameters, while Beijing-based start-up Z.ai released the GLM 4.5 Air model with just 12 billion active parameters.
Alibaba’s Qwen3-Next-80B-A3B is now compact enough to operate on a single Nvidia H200 graphics processing unit, according to Artificial Analysis. On the open-source developer platform Hugging Face, the model quickly broke into the trending leaderboard, amassing almost 20,000 downloads within 24 hours after launch.
Alibaba said its new architecture served as a preview of its next generation of AI models. The future of large language models would likely revolve around refining Alibaba’s approach to address training costs and efficiency, even if entirely different architectures emerge, said Tobias Schroder, an AI researcher at Imperial College London.
“One thing we’ll know for sure is you’re going to have to continually learn … throughout your career,” he said [File]
| Photo Credit: REUTERS
A top Google scientist and 2024 Nobel laureate said Friday that the most important skill for the next generation will be “learning how to learn” to keep pace with change as Artificial Intelligence transforms education and the workplace.
“It’s very hard to predict the future, like 10 years from now, in normal cases. It’s even harder today, given how fast AI is changing, even week by week,” Hassabis told the audience. “The only thing you can say for certain is that huge change is coming.”
The neuroscientist and former chess prodigy said artificial general intelligence — a futuristic vision of machines that are as broadly smart as humans or at least can do many things as well as people can — could arrive within a decade. This, he said, will bring dramatic advances and a possible future of “radical abundance” despite acknowledged risks.
Hassabis emphasised the need for “meta-skills,” such as understanding how to learn and optimising one’s approach to new subjects, alongside traditional disciplines like math, science and humanities.
“One thing we’ll know for sure is you’re going to have to continually learn … throughout your career,” he said.
The DeepMind co-founder, who established the London-based research lab in 2010 before Google acquired it four years later, shared the 2024 Nobel Prize in chemistry for developing AI systems that accurately predict protein folding — a breakthrough for medicine and drug discovery.
Greek Prime Minister Kyriakos Mitsotakis joined Hassabis at the Athens event after discussing ways to expand AI use in government services. Mitsotakis warned that the continued growth of huge tech companies could create great global financial inequality.
“Unless people actually see benefits, personal benefits, to this (AI) revolution, they will tend to become very skeptical,” he said. “And if they see … obscene wealth being created within very few companies, this is a recipe for significant social unrest.”
Mitsotakis thanked Hassabis, whose father is Greek Cypriot, for rescheduling the presentation to avoid conflicting with the European basketball championship semifinal between Greece and Turkey. Greece later lost the game 94-68.
As the smartphone market nears saturation, smart glasses are emerging as the next frontier for AI-enabled wearable devices. Foxconn is positioning itself beyond contract assembly by investing in local augmented reality (AR) technology company Jorjin…
Learn five key areas to target when laying the groundwork for a potential AI implementation at your facility.
Brand Insights from Easy Automation, Inc.
We are in a transformative era, marked by the increasing implementation of AI in both our personal and professional lives. We’ve already seen tools like ChatGPT make their way into our conversations, and we don’t see these new tools going away. While there are still many unknowns surrounding AI and its potential benefits in agricultural facilities, we believe there is a significant opportunity for these new technologies to enhance the efficiency, safety, and profitability of our facilities.
While there are many different levels of comfort and acceptance in implementing AI tools at our facilities, we’ve identified five key areas to target when laying the groundwork for a potential AI implementation at your facility.
Clean and Refine Existing Data
Identify Missing Data and Capture It
Modernize Technology Stack and Storage
Clarify and Enhance Data Security
Align with Forward-Moving Partners
Clean and Refine Existing Data
Where is your data being recorded and stored? How many different software programs or spreadsheets do you have that store your data? Are those individual systems talking to each other, or is there duplicate data? AI technology can only run as efficiently as the data that is provided. In the agricultural facilities we work with, we often see multiple different software programs, including accounting, formulation, order management, trucking, automation, and many others. While many of these programs are necessary for each facility to achieve its business objectives, the systems must work together to provide clean, accurate, and real-time data to be compatible with any future AI integration.
Identify Missing Data and Capture It
Is there an area in your operation where you don’t have any real information or data? Consider your equipment, hazard monitoring sensors, bin levels, truck routing, fleet management, and truck flow within your facility. What comes to mind for your facility? While some new-built facilities capture all this information from the beginning, as our facilities evolve, there are often areas that are missed. Without this data, we are seeing an inaccurate picture of your whole facility from a data standpoint. The power of AI lies in its ability to see the complete picture of data and draw insights and predictions from historical data. Invest in identifying your missing data and take steps to capture it in preparation for future AI implementation.
Modernize Technology Stack and Storage
At a minimum, your facility needs to be connected to the internet, and data must be stored on an accessible platform. Unfortunately, Excel documents on a desktop won’t suffice. Our recommended criteria for modernizing your technology stack include storing in an easily accessible database that offers API connectivity and cloud-based storage. They can log real-time, all-inclusive facility data quickly and accurately. We aim to avoid data silos with multiple disparate data storage areas and prevent systems that are difficult to access or integrate with. API connectivity will be essential, and we want to avoid any systems that require cumbersome custom development to connect to.
Clarify and Enhance Data Security
Security must be at the forefront of the AI implementation conversation. Your data is one of your most valuable assets. We want to ensure that where you place your data or who you allow to analyze it is a reputable source that has been rigorously vetted. Before placing your data in any AI program, it is essential to understand all of the data privacy and security terms and conditions.
Align with Forward-Moving Partners
Do you want to be an expert in AI implementation at your facility? Maybe. However, we recommend aligning yourself with a partner in the industry who is moving forward in that direction and allowing them to become experts, meeting your needs in this area. It is essential to ask questions that provide insight into where that partner is today, as well as where they are headed in the future. Add it to your company’s roadmap and ensure it is also included on your partners’ roadmaps.
At Easy Automation, we have AI implementation on our roadmap and are actively taking steps forward to provide a solution that makes the most sense for our customers. Are you interested in seeing how we might align or learning more about this? Contact our team at 507-728-8214 or by visiting our website at www.easy-automation.com.
Written by Brian Sokoloski – CTO at Easy Automation, Inc.
