AI & Tech · Beijing
Alibaba unveils Qwen4-Max, a closed flagship AI model pitched against OpenAI and Anthropic
The closed-weight model costs 67% less per task than its predecessor, Qwen3.8-Max, and provisionally ranks fifth overall on the Arena.ai leaderboard, behind four US firms.
AI text based on reports by pb.pl, Notícias ao Minuto, EL MUNDO and other publishers. Published . How it is made
For membersThe first 30 seconds for everyone. Synthetic voice (AI).
In brief
Alibaba says Qwen4-Max scored near the top of coding benchmarks, ahead of Anthropic's Claude Fable 5.1 and Moonshot's Kimi K3.1 Pro, and beats its August predecessor Qwen3.8-Max.
Artificial Analysis puts Qwen4-Max at $3.27 per task on its intelligence index, against $10.02 for Qwen3.8-Max.
Arena.ai provisionally ranks Qwen4-Max first among Chinese models and fifth overall, behind Anthropic, Google, OpenAI and Meta.
Qwen4-Max is a closed model with 2.4 trillion parameters, and Alibaba has only promised smaller Qwen versions for release.
Alibaba says it filtered training data to remove offensive cyberattack capabilities while keeping defensive uses such as vulnerability detection.
Launch and benchmark claims
Alibaba presented Qwen4-Max on Monday, 16 November, as its new flagship AI model, positioned against systems from OpenAI and Anthropic. According to Alibaba, the model scored near the top of coding benchmarks and outperformed Anthropic's Claude Fable 5.1 and Moonshot's Kimi K3.1 Pro. The company says it also improves on its predecessor, Qwen3.8-Max, which was unveiled in August, while using fewer tokens. Alibaba describes gains in coding, reasoning and complex professional work, and lists uses that range from 3D modelling and game development to finance, law, healthcare and scientific research. The release is one of a run of Chinese model launches, alongside Tencent's Hunyuan 4 and Moonshot's Kimi K3.1 Pro, from companies that are challenging the lead of US AI firms.
Size, rankings and cost
Qwen4-Max is a closed model with 2.4 trillion parameters in its main architecture, a figure close to the 2.8 trillion attributed to Kimi K3. On the Arena.ai evaluation platform, it was provisionally placed first among Chinese models and fifth overall, behind systems from Anthropic, Google, OpenAI and Meta. Artificial Analysis puts the cost at $3.27 per task on its intelligence index, 67% below the $10.02 for Qwen3.8-Max. The reduction results from both lower prices and lower token consumption, tokens being the units AI systems use to process and generate responses.
- Qwen4-Max
- 2.4 trillion parameters
- Kimi K3
- 2.8 trillion parameters
Data from the article text
- Qwen3.8-Max
- 10.02 USD per task
- Qwen4-Max
- 3.27 USD per task
Data from the article text
Cybersecurity safeguards
Alibaba says it filtered the training data to remove capabilities that could support cyberattacks, while keeping data used for defensive work such as finding and fixing software vulnerabilities. Tests run during training targeted code injection, memory management errors and flawed access controls. The company also says it strengthened protections against the use of AI for cybercrime and for creating dangerous material. Additional post-training measures aim to mitigate chemical, biological, radiological and nuclear risks, according to the company. Alibaba's internal review concluded that the model's cyberdefence capabilities improved substantially and that protection against offensive use was also significantly reinforced. These safety claims come from Alibaba's own statement.
Access and infrastructure plans
Qwen4-Max is not released as an open-weights model, so users cannot download it and modify it themselves. Alibaba has said that smaller versions of the Qwen family will be made available. Bloomberg reports that the Qwen family is the most downloaded open-weights model family in the world, a result it attributes to a large-scale push led by group chief executive Eddie Wu. Alibaba is building a full AI ecosystem that runs from models to its own AI chips. The company aims to reach artificial general intelligence and plans data centres with a combined capacity of 20 GW by 2032.
This article is part of a thread
AI: capabilities, regulation, labour
What changed since this entry:
- 16/11Alibaba unveils Qwen4-Max, 67% cheaper per task than predecessor, fifth on Arena.ai
- 16/11Anchor adds the 15 November US-led AI supply-chain declaration with eight partners
Ask about this text
Sources and context
- pb.plAlibaba presents flagship model Qwen4-Max. China intensifies competition with US AI leadersAlibaba zaprezentowała flagowy model Qwen4-Max. Chiny wzmagają konkurencję z amerykańskimi liderami AI · original in Polish
- Notícias ao MinutoAlibaba reinforces cybersecurity with new AI modelAlibaba reforça cibersegurança com novo modelo de IA · original in Portuguese
- EL MUNDOAlibaba presents Qwen4-Max, its new AI model marked by strict cybersecurity safeguardsAlibaba presenta Qwen4-Max, su nuevo modelo de IA marcado por estrictas salvaguardas de ciberseguridad · original in Spanish
- HandelsblattQwen4-Max: Alibaba introduces new AI flagship modelQwen4-Max: Alibaba stellt neues KI-Flaggschiffmodell vor · original in German
- Eddie Wu
- Group chief executive of Alibaba, credited with the push that built the Qwen model family
- Places
- Beijing, China, United States
- Organizations
- Anthropic, OpenAI, Alibaba Group
Pollar is free and carries no ads in the text. Its members keep it going.
Support Pollar