Alibaba unveils Qwen 3.8, touts chip design and biology capabilities

Alibaba released a promotional video for Qwen 3.8 in which the model spends hours planning chips before moving on to tasks in biology. The timing is not accidental: faced with United States restrictions on exporting accelerators and chip-manufacturing equipment to China, Alibaba presents a model intended to bridge technology gaps without reliance on American hardware. The protein-focused scenes echo AlphaFold and signal an ambition to compete with U.S. companies in scientific discovery. Alibaba has not disclosed the training data for Qwen 3.8, so the corpus and licensing terms remain unknown. The distinction between open weights and open source is relevant: publishing weights allows local inference, but without training data and full training code the model cannot be reproduced or audited. The video does not reveal memory or hardware requirements for running the model, leaving practical questions unanswered. Alibaba has not issued an official technical specification for Qwen 3.8, including parameter count, VRAM needs, or a release date for the weights.