Alibaba released Qwen3.8-2.4T-A95B on August 12, 2026, saying it is the first Qwen-Max-class model to be made available as open weights. It is the company's first open-weight release since the Qwen3.8-Max launch.
Alibaba reported benchmark scores of 93.0 on PaperBench, 82.8 on IFBench, and 86.6 on terminal-based coding, measured on research workflows, instruction following, and coding tasks. That compares with scores from leading frontier models across most categories.
Qwen3.8-2.4T-A95B is built on a hybrid linear-plus-full-attention architecture and targets agentic execution, including multi-step coding, long-horizon planning, and autonomous tool use. Availability begins with deployment on Amazon SageMaker HyperPod, initially for organizations evaluating self-hosted alternatives to proprietary APIs.
"This is the second post in our series on deploying open trillion-parameter models on Amazon SageMaker HyperPod," said the Alibaba Qwen team. The post covers the full path from cluster provisioning to an OpenAI-compatible endpoint.
The announcement follows the first post in the series covering Kimi K3. Alibaba did not say how the model will perform on harder repository-level tasks and general tool use, and noted that these results position Qwen3.8-2.4T-A95B as a credible frontier-class option, particularly for coding agents and research pipelines.
Source: awsml
