Alibaba Unveils Qwen4-Max Artificial Intelligence Model With Multi-Trillion Parameter Scale
Alibaba Group has introduced its flagship Qwen4-Max artificial intelligence model, featuring a 2.4-trillion-parameter architecture designed to handle advanced programming, cybersecurity, and scientific workloads.
- Alibaba launched Qwen4-Max, featuring 2.4 trillion parameters with 71 billion active per token.
- The model scored 75.7 in coding evaluations, outperforming competing systems in internal benchmarks.
- API pricing for cached inputs starts at $0.075 per million tokens across company cloud platforms.

Alibaba Group has officially introduced Qwen4-Max, its newest flagship artificial intelligence system. The release marks an expansion of the company's efforts in the global technology sector, positioning the system against products offered by OpenAI and Anthropic. According to the firm, the platform targets complex professional duties, including software debugging, financial document analysis, game development, and scientific experiment assistance.
Architecture and Efficiency Improvements
The system relies on the newly developed Qwen4 infrastructure, incorporating a total of 2.4 trillion parameters while activating 71 billion parameters per token. This configuration utilizes a Mixture-of-Experts methodology, engaging only specific sections of the network depending on the nature of the workload to lower computing demands. Additionally, the platform provides a context window supporting up to one million tokens and includes native visual processing capabilities.
Company data indicates notable efficiency gains compared to the prior Qwen3.8-Max iteration. Benchmarks demonstrate token reduction figures of 35 percent on Work360 and 49 percent on Terminal-Bench 4.0. Furthermore, average task costs on Work360 dropped by 62 percent, shifting from $4.43 down to $1.67.
Benchmark Performance and Capabilities
In developer evaluations comparing the software against competing options from OpenAI and Anthropic, Alibaba reported that its flagship platform achieved a score of 75.7 in coding assessments. This figure places it ahead of rival releases such as Anthropic Opus 5.5 at 74.2 and OpenAI GPT-6.1-Sol at 75.2, based on internal testing results provided by the organization.
For cybersecurity operations, the system incorporates features meant to locate software vulnerabilities and recommend corrective measures while preventing assistance with unauthorized cyberattacks. Additional filtering mechanisms target chemical, biological, radiological, and nuclear risk domains. During testing, the platform executed a continuous 48-hour research session that identified an experimentally verified coolant molecule for semiconductor manufacturing.
Availability, Pricing, and Ecosystem
The software is accessible through company services including QwenCloud, QwenWork, Qoder, and QwenStudio. Application programming interface pricing for cached inputs begins at $0.075, roughly equivalent to seven rupees, per million tokens. Alongside the flagship release, the company launched Live Delivery to generate early project prototypes and refine them through user feedback iterations, alongside an open-source companion plugin.
While the primary Qwen4-Max platform remains closed-source to protect the proprietary architecture, the organization confirmed that smaller variants within the Qwen4 product family will be distributed as open weights.
- Qwen4-Max utilizes a Mixture-of-Experts architecture with a one-million-token context window.
- Efficiency improvements include a 35 percent reduction in tokens on the Work360 benchmark.
- Smaller models in the Qwen4 family will be released as open source, while the flagship remains closed.
Spotted an error? · Our editorial policy
UP NEXTDadasaheb Phalke International Film Festival Awards 2026: Taylor Swift and James Cameron Among WinnersSwipe more stories · TechnologyQuick cards, a few lines each. Tap to open.More from Technology

Anthropic Researcher Claims AI-Assisted Proof for the Riemann Hypothesis

Grand Theft Auto VI Becomes Largest Entertainment Launch in History

Trump Backs Jensen Huang’s Return to Nvidia Amid Board Resistance and Intelligence Scrutiny
