Alibaba Group Holding Ltd. unveiled its largest artificial intelligence model, Qwen3.8-Max, built on 2.4 trillion parameters. Announced on August 3, 2026, the model matches performance benchmarks of Anthropic’s Fable 5 while driving a 7% surge in Alibaba’s Hong Kong shares.
Alibaba is driving forward in the global artificial intelligence race with the release of its biggest ever AI model, challenging established U.S. leaders. The system, designated as Qwen3.8-Max, incorporates 2.4 trillion parameters to manage complex reasoning, long-horizon projects, and visual tasks. Markets reacted quickly to the August 3 rollout, sending Alibaba’s New York-listed shares up 4.5% and its Hong Kong shares climbing 7%.
Benchmark Performance and the Trillion-Parameter Scale
While parameter scale does not automatically guarantee superior output, it serves as a closely watched metric for computing capacity and data density. Alibaba’s new release approaches the scale of domestic rival Moonshot AI’s Kimi K3, which debuted at the World Artificial Intelligence Conference in Shanghai with 2.8 trillion parameters. Despite the immense size of these models, both companies utilize a mixture-of-experts
design that divides work among specialised parts of the system rather than activating the entire network for every query.
According to reporting from Reuters, Qwen3.8-Max activates only 95 billion parameters at any single time. This selective activation minimizes operational costs and response delays while handling a context window of up to 1 million tokens, enabling the system to ingest thousands of pages of text or extended video feeds.
Challenging Anthropic and Shifting Market Dynamics
Alibaba shared internal benchmark results indicating that Qwen3.8-Max delivers scores comparable or sometimes superior to Anthropic’s Fable 5, an advanced U.S. model that previously faced temporary export restrictions. Independent tracking platforms place the model competitively across multiple leaderboards, ranking second in the Vision Arena and fifth in the Text Arena.
Industry observers note that these rapid releases reflect a broader closing of the technological gap between Chinese and Western developers. Vey-Sern Ling, managing director at Union Bancaire Privée, emphasized that market participants frequently miscalculate the region’s capabilities.
Autonomous Coding and Open-Weight Strategy
Beyond standard text processing, Qwen3.8-Max demonstrates strong capabilities in software engineering and long-horizon execution. Alibaba reported that in internal testing, the model spent 16 days independently building and refining an AI coding tool by writing code, executing tests, correcting errors, and updating its framework with minimal human intervention.
Commercial positioning is also shifting rapidly. While American providers like OpenAI, Anthropic, and Google rely heavily on closed-source ecosystems, Chinese laboratories continue to emphasize open-weight accessibility to build traction among global developers. Alibaba confirmed that it will release the model weights for public download next week. Furthermore, the company established an aggressive pricing structure for API usage, charging US$2 per one million input tokens and US$6 per million outputs.
Financial Impact and Economic Outlook
The August release arrives alongside a wave of competing announcements across China’s technology sector, including DeepSeek’s low-cost V4-Flash model and new video generators from ByteDance and MiniMax Group.
Keep reading