Alibaba Launches Qwen4-Max, Claims 60% Lower Task Costs
By Suad Seferi ·

Key takeaways
- Alibaba has launched Qwen4-Max with 2.4 trillion total parameters, a one-million-token context window and multimodal capabilities.
- The company claims approximately 60% lower average task costs compared with Qwen3.8-Max, based on its proprietary Work360 benchmark. Independent verification is still needed.
- Qwen4-Max is available through Alibaba's cloud services and APIs, but the flagship model is not currently available as an open-weight release.
Alibaba has released Qwen4-Max, its latest flagship artificial intelligence model, claiming substantial improvements in reasoning, coding and operational efficiency. The company says the new system can complete complex tasks at approximately 60% lower average cost than its predecessor, Qwen3.8-Max. Announced on November 16, the model contains 2.4 trillion parameters, supports text, images and video, and offers a context window of one million tokens. It is now available through Alibaba's cloud services and developer platforms, following a limited preview that began on November 11. The release comes as competition among major AI developers increasingly focuses on how much useful work models can complete, rather than simply how large they are. A more efficient model without increasing its size Qwen4-Max uses an upgraded mixture-of-experts (MoE) architecture, which activates only part of the model's parameters when processing information. Qwen4-Max ranked 6th on Code Arena Although the system contains 2.4 trillion parameters, only 71 billion are activated per token. According to Alibaba, that represents a 25% reduction in active parameters compared with Qwen3.8-Max, despite both models having the same total parameter count. This distinction matters because larger AI models generally require substantial computing resources. Reducing the amount of computation required for each request can improve efficiency without necessarily sacrificing performance. Alibaba reports that Qwen4-Max uses 35% fewer tokens in its high-effort reasoning mode and reduces average task costs by approximately 60% on Work360, the company's proprietary benchmark covering 721 specialized professional fields. However, these figures come from Alibaba's own evaluations. Independent testing will be necessary to establish whether similar savings can be achieved across different applications and workloads. Coding, AI agents and professional work Alibaba is positioning Qwen4-Max as a model capable of handling more complex assignments, including software development, research, document analysis and workflows involving multiple tools. The model supports text, image and video understanding, while its one-million-token context window allows developers to submit substantial amounts of information in a single request. It can also perform function calling, use external tools and generate structured outputs, capabilities that are particularly relevant for AI agents designed to complete tasks across different software systems. Alibaba reports that Qwen4-Max ranked sixth on Code Arena, a leaderboard evaluating coding capabilities. The company also published results from benchmarks examining agent workflows, workplace productivity and visual reasoning. These results suggest improvements, but benchmark rankings alone do not establish which model performs best in real business environments. Reliability, execution costs, response quality and integration requirements remain important factors when choosing an AI system. Alibaba is also targeting scientific research One of the more ambitious claims concerns the model's ability to support extended scientific workflows. Qwen4-Max outperforms Qwen3.8-Max overall on Work360 Alibaba says Qwen4 completed a continuous 48-hour process involving molecular design, testing and iterative improvements to identify a coolant molecule for advanced semiconductor applications. The company reports that the resulting molecule was experimentally validated. In another experiment involving Alzheimer's research, Alibaba says the model executed hundreds of tool calls and thousands of evaluations while designing and optimizing candidate molecules. These examples illustrate the company's ambitions for AI systems that can participate in lengthy research processes rather than simply answer individual questions. They should nevertheless be treated as company-reported demonstrations, not independent evidence that the model can reliably conduct scientific discovery without human supervision. Competitive pricing, but a different approach to open AI Qwen4-Max is available through QwenCloud, QwenWork and QwenStudio, with API access for developers. Alibaba Cloud lists standard pricing for its Singapore deployment at $1.50 per million input tokens and $4.50 per million output tokens. The model also supports API interfaces compatible with OpenAI and Anthropic formats, potentially reducing the engineering work required to test it within existing applications. There is, however, an important difference from Alibaba's earlier open-weight releases. Qwen4-Max is currently distributed as a hosted model rather than being available for developers to download and deploy independently. Alibaba says smaller models from the Qwen4 family will be released with open weights, although it has not announced a timetable. That distinction matters for organizations that want direct control over model deployment, infrastructure and data processing. For software companies and startups in the Balkans, the release provides another option when evaluating AI services for business applications. Competitive pricing and API compatibility could make experimentation more accessible, but organizations handling sensitive information will still need to assess hosting locations, contractual protections and data-processing requirements. The next test is performance outside Alibaba's benchmarks Qwen4-Max represents an attempt to improve AI capability without increasing the total size of the underlying model. The more interesting development is its reported efficiency. If Alibaba's cost and performance claims hold up under independent evaluation, the model could become a competitive option for developers building applications that require extensive reasoning, coding and automated workflows. But the company has not yet established that its reported savings will translate consistently into production environments. The next developments to watch are independent performance…
Frequently asked questions
What is Alibaba Qwen4-Max?
Qwen4-Max is Alibaba's flagship multimodal AI model designed for reasoning, coding, visual understanding and complex workflows involving AI agents.
Is Qwen4-Max more powerful than ChatGPT or Claude?
Alibaba reports competitive benchmark results, but there is insufficient independent evidence to conclude that Qwen4-Max consistently outperforms leading OpenAI or Anthropic models across real-world applications.
Is Qwen4-Max free or open source?
The flagship model is offered through hosted services and APIs. Alibaba has announced plans to release smaller open-weight Qwen4 models, but no release date has been confirmed.
Can developers in the Balkans use Qwen4-Max?
Alibaba documents international API availability, including deployments in Singapore and Germany. Developers should verify account eligibility, regional availability, pricing and data-processing requirements before adopting the service.