Alibaba releases open‑weight Qwen3.8‑2.4T; AWS shows SageMaker HyperPod deployment
Alibaba’s Qwen3.8‑2.4T‑A95B model is now open‑weight, and AWS provides a guide to run it on SageMaker HyperPod with vLLM and NVFP4 quantization.
Original source published: September 9, 2026
On August 12, 2026 Alibaba’s Qwen team released Qwen3.8‑2.4T‑A95B, the first open‑weight model in the Qwen‑Max class. It has 2.4 trillion total parameters, 95 billion activated per token, a hybrid linear‑plus‑full‑attention design, and a native 262 K‑token context that can be extended to 1 M tokens. The model targets agentic workloads such as multi‑step coding, long‑horizon planning and autonomous tool use.
AWS shows how to serve the model on Amazon SageMaker HyperPod with vLLM on an ml.p6‑b300 instance that contains eight NVIDIA B300 Blackwell Ultra GPUs. Using NVFP4 (W4A4) quantization reduces the weight footprint to about 1.2 TB, fitting within the node’s 2.1 TB GPU memory. The HyperPod Inference Operator automates model download, container scheduling, health checks and autoscaling, while vLLM flags enable tensor‑parallelism, prefix caching, reasoning parsing and native Multi‑Token Prediction.