BNB $591.25 +0.37%
XRP $0.959657 +0.92%
ETH $1,826.27 +1.02%
BTC $64,547.14 +0.32%
BNB $591.25 +0.37%
XRP $0.959657 +0.92%
ETH $1,826.27 +1.02%
BTC $64,547.14 +0.32%
BREAKING
Technology

Alibaba 2.4-Trillion Qwen 4 Max Tops Chinese Agent Rankings

Alibaba's 2.4-Trillion-Parameter Qwen 4 Max Beats Chinese Rivals in Agent Race

Community Trust ScoreVerified

88%
Real
Verified26 votes
Updated 7 minutes ago

Alibaba released Qwen 4 Max on Monday, a 2.4-trillion-parameter artificial intelligence model that currently holds the top position among Chinese entries on major agent leaderboards. The model represents the first flagship built on Alibaba’s new Qwen 4 architecture and operates with closed weights, limiting direct access for developers.

The release positions Alibaba ahead of domestic competitors including Tencent, Moonshot AI, and DeepSeek in specific agent performance metrics. While American laboratories maintain the top spots in overall agent and web-development tables, Qwen 4 Max has secured the highest ranking for a Chinese model on Arena’s agent leaderboard.

Architectural Shift and Performance Metrics

Qwen 4 Max utilizes a mixture-of-experts design, activating only 71 billion parameters at a time to process inputs. This is a reduction from the previous flagship, Qwen 3.8 Max, which activated 95 billion parameters. The model accepts up to one million tokens of context and can process text, images, and video.

Advertisement

On Agent Arena, the model placed 14th among individual models, while Alibaba ranked as the fifth laboratory overall. The leaderboard evaluates performance based on real agent sessions, considering task completion, user satisfaction, and response to corrections. Qwen 4 Max posted a preliminary score of 7.46% across 14,727 sessions, the highest score among Chinese entries on that specific board.

In coding evaluations, Qwen 4 Max debuted sixth globally on Code Arena with 1,750 points. This figure represents an increase of 74 points from its predecessor. The model also placed eighth in categories focused on branding and reference-based design. Alibaba’s demonstrations show the model building game environments and translating factory layouts into 3D models. The visual capabilities are designed for understanding images and video to produce code or text-based outputs, rather than generating new images.

Cost Reductions and Access Restrictions

Alibaba has lowered the cost of using Qwen 4 Max compared to its previous flagship. Through QwenCloud, the model costs $1.50 per million input tokens and $4.50 per million output tokens. This represents a 25% reduction from the previous rates of $2 per million input tokens and $6 per million output tokens. Reused input served from an automatic cache costs $0.075 per million tokens.

Internal testing by Alibaba indicated that the model used 35% fewer tokens compared with Qwen 3.8 Max. The company reported that average task costs were reduced by roughly 60% in comparison to the older model. These efficiency gains aim to make the model more competitive in the API market, where DeepSeek has also been competing on price.

Unlike its predecessor, Qwen 4 Max is closed-weight. Developers cannot download, modify, or run the model privately on their own hardware. Access is available only through an API or Alibaba’s applications. This stands in contrast to the Qwen 3.8 Max, which Alibaba opened earlier this year. The shift to closed weights for the flagship model limits the ability for independent developers to fine-tune or deploy the system locally.

Ecosystem Impact and Future Releases

The Qwen ecosystem has seen significant growth on Hugging Face. An August report from the platform counted 151,448 Qwen-based derivative models. This number is approximately 4.7 times the number of models built specifically on Meta’s Llama family. The large number of derivatives highlights the model’s influence in the open-source community, even as the new flagship restricts direct access.

Alibaba previewed elements of the Qwen 4 architecture with smaller models months before this release. The company has stated that smaller Qwen 4 models will be released with open weights in the future. However, Alibaba has not announced the specific sizes or release dates for these open-weight models. The current flagship remains exclusive to API users, maintaining Alibaba’s control over the highest-performing version of its technology.

Community Trust IndexHigh Confidence
88%
Real
Real88%12%Fake
26 community signals

Evie Vavasseur

Evie Vavasseur is a crypto writer and digital content specialist covering the latest developments in blockchain technology, decentralized finance, and the broader digital asset ecosystem. With a keen eye for emerging trends, Evie provides accessible and insightful coverage of cryptocurrency markets, NFTs, and Web3 innovations for The Currency Analytics.

Advertisement

Related Stories