Leer más

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.

This article originally appeared in the South China Morning Post (SCMP), the most authoritative voice reporting on China and Asia for more than a century. For more SCMP stories, please explore the SCMP app or visit the SCMP’s Facebook and Twitter pages. Copyright © 2025 South China Morning Post Publishers Ltd. All rights reserved.

Copyright (c) 2025. South China Morning Post Publishers Ltd. All rights reserved.

At an open-air summit in Athens, Demis Hassabis, head of Google’s DeepMind and Nobel chemistry laureate, argued that the skill most needed in the years ahead will be the ability to keep learning. He described education as moving into a period where adaptability matters more than fixed knowledge, because the speed of artificial intelligence research is shortening the lifespan of expertise.

Hassabis said future workers will have to treat learning as a constant process, not a stage that ends with graduation. He pointed to rapid advances in computing and biology as examples of how quickly fields now change once AI tools enter the picture.

Outlook on technology

The DeepMind chief warned that artificial general intelligence may not be far away. In his view, it could emerge within a decade, carrying a weight of opportunity and risk. He described its potential impact as larger and faster than the industrial revolution, a shift that could deliver breakthroughs in medicine, clean energy, and space exploration.

Even so, he stressed that powerful models must be tested carefully before being widely deployed. The practice of pushing products out quickly, common in earlier technology waves, should not guide the release of systems capable of influencing economies and societies on a global scale.

Prime minister’s caution

Greek Prime Minister Kyriakos Mitsotakis, who shared the stage at the Odeon of Herodes Atticus, said governments will struggle to keep pace with corporate growth unless they adopt a more active role. He warned that when the benefits of technology are concentrated among a small set of companies, public confidence erodes. He tied the issue to social stability, saying communities won’t support AI unless they see its value in everyday life.

Mitsotakis pointed to Greece’s efforts to build an “AI factory” around a new supercomputer in Lavrio. He presented the project as part of a wider European push to turn regulation and research into competitive advantages, while reducing reliance on U.S. and Chinese platforms.

Education and jobs

Both speakers returned repeatedly to the theme of skills. Hassabis said that in addition to traditional training in science and mathematics, students should learn how to monitor their own progress and adjust their methods. He argued that the most valuable opportunities often appear where two fields overlap, and that AI can serve as a tutor to help learners explore those connections.

Mitsotakis said the challenge for governments is to match school systems with shifting labor markets. He noted that Greece is mainly a service economy, which may delay some of the disruption already visible in manufacturing-heavy nations. But he cautioned that job losses are unavoidable, including in sectors long thought resistant to automation.

Strains on democracy

The prime minister voiced concern that misinformation powered by AI could undermine elections. He mentioned deepfakes as a direct threat to public trust and said Europe may need stricter rules on content distribution. He also highlighted risks to mental health among teenagers exposed to endless scrolling and algorithm-driven feeds.

Hassabis agreed that lessons from social media should inform current choices. He suggested AI might help by filtering information in ways that broaden debate instead of narrowing it. He described a future where personal assistants act in the interest of individual users, steering them toward content that supports healthier dialogue.

The question of abundance

Discussion also touched on the idea that AI could usher in an era of radical abundance. Hassabis said research in protein science, energy, and material design already shows how quickly knowledge is expanding. He argued that the technology could open access to vast resources, but he added that how wealth is shared will depend on governments and economic policy, not algorithms.

Mitsotakis drew parallels with earlier industrial shifts, warning that if productivity gains are captured only by large firms, pension systems and social programs will face heavy strain. He said policymakers must prepare for a period of disruption that could arrive faster than many expect.

Greece’s role

The Athens event also highlighted the country’s ambition to build a regional hub for technology. Mitsotakis praised the growth of local startups and said incentives, venture capital, and government adoption of AI in public services would be central to maintaining momentum.

Hassabis, whose family has roots in Cyprus, said Europe needs to remain at the frontier of AI research if it wants influence in setting ethical and technical standards. He called Greece’s combination of history and new infrastructure a symbolic setting for conversations on the future of technology.

Preparing for the next era

The dialogue closed on a shared message: societies will need citizens who can adapt and learn throughout their lives. For Hassabis, this adaptability is the foundation for navigating a future shaped by artificial intelligence. For Mitsotakis, the task is making sure those changes strengthen democratic values rather than weaken them.

Notes: This post was edited/created using GenAI tools.

Read next: Most Americans Now Rely on AI in Search and Shopping, Survey Finds