The expansion will grow its existing network of 91 facilities across 29 regions and places the Chinese company head-to-head with established giants like Amazon Web Services, Microsoft Azure, and Google Cloud.

Earlier this year, Alibaba pledged ¥380 billion ($53.4 billion) over three years to build AI infrastructure. At its annual Apsara Conference, Chief Executive Eddie Wu signalled the spending will not stop there. “The speed of AI industry development has far exceeded our expectations, and the industry’s demand for AI infrastructure has also far exceeded our expectations,” he told participants.

The highlight of the event was the unveiling of Qwen3-Max, Alibaba’s largest artificial intelligence model so far. Built with more than 1 trillion parameters, it has been designed to handle code generation and operate as an autonomous agent, systems that can pursue goals with limited human direction.

According to Alibaba Cloud’s Chief Technology Officer, Zhou Jingren, the model displayed particular strength in tasks that typically require continuous prompts. Independent benchmarks such as Tau2-Bench showed it surpassed rival offerings including Anthropic’s Claude and DeepSeek-V3.1 in some areas.

Alibaba also presented Qwen3-Omni, a multimodal system developed for immersive applications ranging from smart glasses to intelligent cockpits. This is a drive into consumer-facing AI experiences, particularly in retail, automotive, and wearable technology.

In a further step, the company revealed a partnership with Nvidia to advance physical AI capabilities. Their collaboration will focus on model training, data synthesis, reinforcement learning, and validation for real-world use cases, areas where Nvidia dominates the global chip market.

Alibaba, once known primarily for e-commerce, is now placing AI and cloud services at the centre of its global operations, strengthening a competition that stretches across Asia, Europe, and the Americas.

The new model is now available on Hugging Face, an AI development platform, where it has already gained attention for its improved reasoning and coding capabilities. 

DeepSeek claims that this version surpasses the previous model in multiple benchmarks, particularly in mathematical problem-solving and software development.

One of the improvements is its performance on the American Invitational Mathematics Examination (AIME), where it scored 59.4, a notable jump from the previous model’s 39.6. 

Similarly, on LiveCodeBench, a coding assessment, it gained 10 points to reach 49.2. These improvements suggest a more capable AI system for both research and practical applications.

With 685 billion parameters, DeepSeek-V3-0324 slightly surpasses the earlier V3 model’s 671 billion. Unlike old models, which used a proprietary commercial license, the latest version is distributed under the MIT license, making it more accessible to developers worldwide.

DeepSeek’s development has caught the attention of experts. “Anthropic and OpenAI are in trouble,” said Kuittinen Petri, a lecturer at Häme University of Applied Sciences, on X.

He tested the model by instructing it to “create a great-looking responsive front page for an AI company,” and it successfully generated a fully functional, mobile-friendly website with 958 lines of code.

Apple research scientist Awni Hannun ran the model on a 512GB M3 Ultra workstation. While it processed over 20 tokens per second, he noted that memory usage peaked at 381GB, which, though high, remained within expectations for a model of this scale.

DeepSeek’s progress in AI development has been commendably fast. After launching its V3 model in December and the R1 model in January, speculation is already building about the release of R2, a possible follow-up to its reasoning-focused series. “The coding capabilities are much stronger, and the new version may pave the way for the launch of R2,” said Li Bangzhu, founder of AIcpb.com.

Jasper Zhang, a University of California, Berkeley graduate and maths Olympiad gold medallist, also tested the model using an AIME 2025 problem. “It solved it smoothly,” he noted, asserting that DeepSeek’s models are closing the gap with their Western competitors.

Fahd Mirza, lead cloud and AI engineer at Australian construction materials company Boral, described DeepSeek-V3-0324 as “mind-blowing.” On his YouTube channel, he shared a demonstration of the model tackling complex coding and mathematical tasks, calling its performance “outstanding.”

DeepSeek’s approach to AI development has focused on efficiency and accessibility. Unlike heavily funded rivals, it operates with significantly fewer financial resources. Petri pointed out, “DeepSeek is doing all this with just [roughly] 2 per cent [of the] money resources of OpenAI.”

The innovative technology is aimed at professionals in the music, film, and video game industries, providing tools to generate original sounds and alter existing audio recordings.

Fugatto, an acronym for Foundational Generative Audio Transformer Opus 1, leverages advanced AI capabilities to create sounds based on text prompts.

It also has the ability to modify existing audio, such as converting piano notes into a human-sung melody or altering the emotional tone and accent in recorded speech. This dual functionality distinguishes it from other generative AI tools available today.

Unlike similar technologies developed by companies like Meta Platforms or emerging startups such as Runway, Nvidia’s Fugatto offers features targeting professionals. 

For instance, it can produce imaginative soundscapes—such as a trumpet imitating a barking dog—or craft dynamic audio transitions, like a shift from a thunderstorm to a serene dawn.

Bryan Catanzaro, Nvidia’s vice president of Applied Deep Learning Research, revealed how generative AI could redefine audio production. “Over the past 50 years, computers and synthesizers have significantly changed how music sounds. Generative AI now brings even greater potential to music, gaming, and creative projects for everyone.”

Nonetheless, Nvidia has cautioned regarding Fugatto’s public release. The model, trained on open-source audio data, carries ethical risks, including potential misuse for generating misinformation or violating copyright laws. 

“Any generative technology carries risks. We need to be cautious, which is why we don’t plan to release this immediately,” Catanzaro explained.

Concerns over generative AI misuse have prompted Nvidia and other developers to carefully evaluate safeguards before public deployment. The industry is still facing challenges such as unauthorised imitation of protected content and potential legal issues.

While Nvidia’s Fugatto is still unavailable for public use, the announcement comes as generative AI tools for creative industries are gaining more interest. These technologies are seen as important for enhancing content personalisation and offering new possibilities in advertising, education, and entertainment. 

Earlier this year, Nvidia briefly surpassed Apple in market valuation, achieving a peak of $3.53 trillion. Its success has been driven by strong demand for advanced chips used in AI applications. 

Nvidia’s influence in the sector was further strengthened by its partnership with OpenAI, whose ChatGPT relies heavily on Nvidia GPUs for training.

Positioned as an alternative to more expensive and resource-intensive models, GPT-4o Mini will replace GPT-3.5 Turbo, bringing improved performance at a fraction of the cost.

GPT-4o Mini is designed to scale in reasoning tasks involving both text and vision, making it a versatile tool for a wide range of applications.

According to OpenAI, the model scores 82% on the Massive Multitask Language Understanding (MMLU) benchmark, surpassing the initial and other leading small models in the industry, such as Gemini 1.5 Flash and Claude 3 Haiku.

Olivier Godemont, OpenAI’s Head of Product API, pointed to the importance of affordability in AI accessibility. “For AI to truly empower every corner of the world, we need to make our models more affordable. GPT-4o Mini is a significant step in that direction,” he stated in an interview.

The new model is far more affordable to run than others, costing over 60% less than GPT-3.5 Turbo. For developers, the pricing structure is highly competitive, with the new model priced at 15 cents per million input tokens and 60 cents per million output tokens.

This makes it an attractive option for high-volume, simple tasks that require repeated AI calls.

Developers will find the model’s context window of 128,000 tokens, roughly the length of a book, particularly useful for extensive applications. OpenAI also hinted at future updates that will incorporate video and audio capabilities, further broadening the model’s utility.

Despite its smaller size, the new model is lauded as being faster, smarter, and more cost-efficient than other small models like Llama 3 8b, Claude Haiku, and Gemini 1.5 Flash. The model was rigorously tested on the LMSYS.org chatbot arena, ensuring its competitive edge in the market.

OpenAI has already partnered with companies like Ramp and Superhuman to test GPT-4o Mini. Ramp used the model to develop a tool that extracts expense data from receipts, while Superhuman integrated it into their email client for auto-suggestion features.

These early applications show that the model can simplify tasks and improve productivity in various domains.

Starting today, GPT-4o Mini is available to users of the ChatGPT web and mobile app, with enterprise access rolling out next week.