Skip to content
Market MoodExtreme Fear11
Market Cap: $2.22T −2.43%24h Vol: $76.90BBTC: $64,715 +0.23%
NewsMarketsBitcoinEthereumAltcoinsAsiaWeb3EventsAbout Us
LanguageENRU
Subscribe Pro
Asia

Alibaba launches Qwen 4 Max, China’s biggest AI model with 2.4 trillion parameters

Alibaba unveiled Qwen 4 Max, a 2.4‑trillion‑parameter model that tops Chinese rankings on Arena and offers visual‑aware AI services via API.

· 3 min read
A traditional junk boat with red sails on Victoria Harbour in Hong Kong at night
Photo by Andres Garcia on Unsplash

Key takeaways

  • Qwen 4 Max has 2.4 trillion parameters and a one‑million‑token context window.
  • The model costs $1.50 per million input tokens and $4.50 per million output tokens on QwenCloud.
  • It leads Chinese entries on Arena’s agent leaderboard with a 7.46 % score across 14,727 sessions.

Alibaba Group announced the release of Qwen 4 Max on Monday, positioning it as the company’s most powerful foundation model to date. With 2.4 trillion parameters and a one‑million‑token context window, the model can process text, images and video, and is the highest‑ranked Chinese entry on the Arena AI‑agent leaderboard. The launch marks a new milestone in China’s race to build large‑scale, multimodal AI systems, while the model remains closed‑weight and accessible only through Alibaba’s cloud API.

Performance and ranking

According to Decrypt, Qwen 4 Max posted a preliminary 7.46 % score across 14,727 agent sessions on Arena, the best result among Chinese competitors and placing the model 14th among all individual models worldwide. In the Code Arena web‑development ranking, the model debuted at sixth globally with 1,750 points, a 74‑point jump from its predecessor Qwen 3.8 Max. The improvements are attributed to a mixture‑of‑experts architecture that activates only 71 billion parameters per token, reducing compute demand relative to earlier versions.

Pricing and accessibility

Alibaba priced Qwen 4 Max at $1.50 per million input tokens and $4.50 per million output tokens via the QwenCloud service, a 25 % discount from the previous flagship model. Reused input from an automatic cache is billed at $0.075 per million tokens, making repetitive workflows considerably cheaper. Internal testing cited by the company shows a 35 % reduction in token usage and roughly 60 % lower average task costs compared with Qwen 3.8 Max. Despite these cost advantages, the model’s weights are closed, meaning developers can only call the model through an API or Alibaba‑hosted apps and cannot download or fine‑tune the core parameters.

Implications for China’s AI ecosystem

The release arrives amid a crowded Chinese AI landscape. Moonshot AI’s Kimi K3 and DeepSeek’s models have been pushing open‑weight performance, while American labs continue to dominate the top spots on Arena’s overall tables. Alibaba’s decision to keep Qwen 4 Max proprietary contrasts with its earlier move to open the weights of Qwen 3.8 Max and signals a hybrid strategy: offering a premium, closed‑weight service while promising smaller, open‑weight Qwen 4 variants in the future. Hugging Face data shows more than 150,000 Qwen‑based derivative models on its platform, underscoring the broader community’s reliance on Alibaba’s earlier releases.

What to watch next

Analysts will monitor whether Alibaba’s smaller, open‑weight Qwen 4 models arrive before the end of 2027, and how pricing compares with DeepSeek’s increasingly aggressive API rates. The model’s visual‑understanding capabilities, demonstrated in demos that translate factory layouts into 3D scenes, could also influence Chinese enterprise adoption in manufacturing and design. Finally, the ongoing performance gap between Chinese and U.S. models on Arena may narrow if Alibaba expands its open‑weight ecosystem or if regulatory shifts affect cross‑border AI collaboration.

Frequently asked questions

Can developers download or fine‑tune Qwen 4 Max?

No. The model is closed‑weight and can only be accessed through Alibaba’s API or cloud services.

How does Qwen 4 Max’s cost compare with its predecessor?

Input tokens cost $1.50 per million (down from $2) and output tokens $4.50 per million (down from $6), a 25 % reduction.

What makes Qwen 4 Max stand out among Chinese AI models?

It leads Chinese entries on Arena’s agent leaderboard with a 7.46 % score and offers a one‑million‑token context window plus native visual understanding.

Sources

  • Tokenpost: Alibaba's Qwen4-Max Scores 43, 16 Points Behind Claude Opus 5.5
  • Decrypt: Alibaba's Qwen 4 Max Leads China's AI Agent Race—But You Can't Download It

Not financial advice. Crypto assets are volatile and you can lose money. Read our Financial Risk Disclosure.

AlibabaQwen 4 MaxAIChinaArena leaderboardRegulationInstitutions
Newsletter

Get the BitcoinEast Daily

Top crypto news, analysis, and market insights, delivered every day.