Alibaba says new AI model outperforms OpenAI, Anthropic's top offerings
What's the story
Alibaba has launched its new flagship artificial intelligence (AI) model, the Qwen4-Max. The e-commerce giant claims that this new model can compete with the likes of Anthropic and OpenAI. The announcement comes as a major step in China's challenge to US tech giants in the AI space. The Qwen4-Max has performed exceptionally well in coding benchmarks, outperforming both Anthropic's Claude Fable 5.1 and Moonshot's Kimi K3.1 Pro.
Model details
Qwen4-Max is a closed-source model
Unlike some of Alibaba's previous models, Qwen4-Max is a closed-source model. This means developers can't download its underlying weights and run it independently.
However, the company has promised that smaller models in the Qwen4 family will be released as open source.
The new model comes with improved capabilities in coding, cybersecurity, professional work, and scientific research.
It can perform complex tasks like writing/debugging code, building games, analyzing financial documents, and assisting in scientific experiments.
Technical specs
It has a 1-million-token context window
Qwen4-Max is built on Alibaba's new Qwen4 architecture and has a 2.4-trillion-parameter backbone, with 71 billion parameters activated for each token processed.
This Mixture-of-Experts approach allows the model to handle demanding workloads while keeping computing requirements lower.
The model also supports a one-million-token context window, allowing it to process large amounts of information in a single conversation.
It comes with native vision capabilities, which means it can understand visual information alongside text.
Performance metrics
It is also more efficient than its predecessor
Alibaba claims that Qwen4-Max is more efficient than its predecessor, the Qwen3.8-Max model.
It uses up to 35% fewer tokens on its Work360 benchmark and 49% fewer on Terminal-Bench 4.0.
The company also reports a whopping 62% reduction in average spending per task on Work360, from $4.43 to $1.67.
These improvements make Qwen4-Max a cost-effective option for businesses looking to leverage AI technology for their operations.
Benchmark results
Qwen4-Max can handle cybersecurity tasks better
In internal tests, Alibaba found that Qwen4-Max outperformed other leading models such as OpenAI's GPT-6.1-Sol and GPT-6 Astra, Anthropic's Opus 5.5, and Kimi K3.1 on several benchmarks.
The company says the new model is better at handling cybersecurity tasks too.
It can detect software vulnerabilities and recommend patches while preventing information that could be used for cyberattacks from being leaked in the first place.
Access details
Availability and pricing of the new model
Qwen4-Max is available through Alibaba's QwenCloud, QwenWork, Qoder and QwenStudio platforms.
The company has also launched Live Delivery, a feature that lets the model deliver an early working version of a project and improve it through subsequent updates based on user feedback.
The cached-input API pricing for the model starts at 0.5 yuan per million tokens.
Alibaba has also promised to release smaller open-source models in the future.