Alibaba Unveils Qwen3.8-Max AI Model Featuring 2.4 Trillion Parameters
Alibaba Group announced Qwen3.8-Max, its newest artificial intelligence model built with 2.4 trillion parameters. The model processes text, images, video, and complex documents in a single system.

The system relies on a Mixture-of-Experts architecture. This design routes incoming requests through specialized sub-networks rather than activating the entire network at once.
Only 95 billion parameters run during any single query. This partial activation lowers response delays and cuts computing expenses for enterprise users.
Benchmark Performance and Technical Specs
Alibaba submitted Qwen3.8-Max to the crowdsourced evaluation platform Arena.AI. The model scored as the top-performing Chinese text model, trailing globally behind Anthropic's Claude Fable 5.
In visual analysis tasks, Qwen3.8-Max claimed the second position globally, ranking behind an Anthropic variant.
The model supports a context window of 1 million tokens. Developers can input entire software codebases, legal archives, or long manuals in one request.
According to Alibaba's technical team:
"Qwen3.8-Max demonstrates proficiency in autonomous coding and long-horizon execution, operating independently over extended periods without human intervention."
In internal demonstrations, the model created a self-evolving engineering framework from scratch by running an automated user feedback loop.
Market Dynamics and Industry Competition
The launch expands on the previous Qwen 3.7-Max release. The earlier model focused on agent workflows, and Qwen3.8-Max adds visual processing to those automated capabilities.
The announcement follows the release of Moonshot AI's Kimi K3 model, which contains 2.8 trillion parameters. Chinese developers continue releasing large-parameter models to build open-source developer communities.
Western AI firms such as OpenAI, Google, and Anthropic keep parameter counts undisclosed for closed-source models. Chinese tech firms publish parameter metrics to demonstrate computational scale.
Regarding global market placement, Alibaba stated:
"second only to Claude Fable 5"
Availability and Access Terms
Alibaba launched preview access to Qwen3.8-Max through its Token Plan subscription and its Qoder and QoderWork platforms. Preview users receive a discounted price set at 10% of standard operational rates.
The service supports API endpoints compatible with OpenAI and Anthropic format specifications. This structure permits engineering teams to swap existing model calls into Qwen endpoints.
Alibaba promised an open-weight release for developers in an upcoming update. The company has not published the official download repository, exact release date, or software license terms.