Alibaba and DeepSeek Take Different Paths as China's AI Competition Enters a New Phase

Alibaba has introduced Qwen3.8-Max, its largest artificial intelligence model yet, while DeepSeek has launched the ultra-low-cost V4-Flash, signalling that the next stage of the AI race may be driven as much by affordability and efficiency as by raw computing power.

The announcements highlight two complementary trends shaping the AI industry. Alibaba is focusing on building larger, more capable multimodal systems for complex enterprise tasks, while DeepSeek is positioning itself as a provider of highly affordable AI for large-scale commercial deployment.

Both companies continue to embrace open-weight AI models, a strategy that increasingly distinguishes Chinese developers from many leading Western AI firms.

Quick Facts

  • Alibaba unveiled Qwen3.8-Max, a 2.4 trillion-parameter AI model.
  • The model supports text, images, and video.
  • Qwen3.8-Max processes up to 1 million tokens in a single context window.
  • DeepSeek introduced V4-Flash, one of the lowest-cost large language models according to Artificial Analysis.
  • Alibaba's Hong Kong-listed shares gained about 7% following the announcement.
  • Both companies continue supporting open-weight AI, allowing developers to download and customise model weights.

What Makes Qwen3.8-Max Alibaba's Most Advanced AI Model?

Alibaba describes Qwen3.8-Max as its largest AI system to date, featuring 2.4 trillion parameters, placing it among the biggest AI models developed by a Chinese company. The model approaches the scale of Moonshot AI's Kimi K3, which has 2.8 trillion parameters.

Rather than processing only text, Qwen3.8-Max is designed as a multimodal AI system capable of understanding text, images, and video. The model can also analyse extremely large inputs thanks to its 1 million-token context window, making it suitable for reviewing lengthy legal documents, research papers, software repositories, and other extensive datasets.

Alibaba also said internal testing showed the model completed a software engineering project over 16 days, demonstrating its ability to support long-running development tasks.

The company plans to make Qwen3.8-Max publicly available next week.

Why Does the Mixture-of-Experts Architecture Matter?

Direct Answer

A mixture-of-experts (MoE) architecture improves AI efficiency by activating only a small portion of a model's total parameters for each request instead of using the entire model. This reduces computing costs while maintaining strong performance.

Although Qwen3.8-Max contains 2.4 trillion parameters, Alibaba said only about 95 billion parameters are activated for any individual task.

This selective processing allows the model to deliver advanced capabilities without requiring the immense computational resources that would be needed if every parameter were used simultaneously.

For businesses deploying AI at scale, this approach can translate into lower infrastructure costs and faster responses.

DeepSeek Is Competing on Cost Instead of Size

While Alibaba is emphasising scale and capability, DeepSeek is targeting one of the industry's fastest-growing priorities: AI operating costs.

Research firm Artificial Analysis ranked DeepSeek's V4-Flash as the lowest-cost model among major AI systems evaluated in its benchmark tests.

DeepSeek prices the model at:

  • $0.14 per million input tokens
  • $0.28 per million output tokens

Artificial Analysis estimated that V4-Flash costs roughly $0.03 to complete an average benchmark task.

For comparison:

  • Moonshot AI's Kimi K3: $0.86
  • OpenAI's GPT-5.6 Sol: $1.86
  • Anthropic's Claude Fable 5: $3.15

Researchers noted that benchmark-based cost comparisons provide a more realistic picture of deployment expenses because they account for the computation required to complete real tasks rather than published pricing alone.

Why Open-Weight AI Is Becoming a Competitive Advantage

One of the biggest differences between Alibaba and DeepSeek compared with companies such as OpenAI, Anthropic, and Google is their continued investment in open-weight AI models.

Open-weight models allow developers to download trained model weights and adapt them for specific applications. This gives organisations greater flexibility to customise AI systems, host them on private infrastructure, and reduce dependence on proprietary cloud platforms.

Many enterprises also value the increased transparency and control offered by open-weight models, particularly in industries with strict security or compliance requirements.

According to Omdia analyst Lian Jye Su, organisations increasingly prioritise AI systems that are affordable, transparent, and capable enough for practical business use, rather than simply selecting the highest-performing model available.

 

Why AI Cost Is Becoming as Important as AI Intelligence

For much of the recent AI boom, companies competed by building larger models and achieving higher benchmark scores.

That competitive landscape is beginning to evolve.

As enterprises integrate AI into customer service, software development, data analysis, and business operations, recurring inference costs have become a major consideration. Even modest cost differences can translate into substantial savings when millions of AI requests are processed every day.

Alibaba's emphasis on efficient model architecture and DeepSeek's focus on ultra-low inference costs reflect this shift toward commercial sustainability.

Instead of asking only, "Which model is the smartest?" businesses are increasingly asking, "Which model delivers the best value at scale?"

 

What Comes Next?

Alibaba's release of Qwen3.8-Max and DeepSeek's introduction of V4-Flash underscore the rapid pace of China's AI development.

One company is pushing the boundaries of multimodal capability and large-context reasoning, while the other is redefining affordability for enterprise AI deployments.

As businesses weigh performance, transparency, flexibility, and long-term operating costs, competition among AI developers is likely to extend well beyond benchmark rankings.

The next phase of the AI industry may ultimately be decided not only by which model performs best, but also by which one organisations can deploy most efficiently.

Why It Matters

The latest announcements from Alibaba and DeepSeek highlight a broader shift in artificial intelligence. Success is no longer determined solely by building the largest model. Efficiency, deployment costs, openness, and real-world usability are becoming equally important factors. As enterprises adopt AI across more business functions, models that combine strong performance with lower operating costs could gain a significant competitive advantage in the global market.