Free perks leveled up! Qwen3.8-Flash is now live on http://B.AI, fully open for free access! As a preview of the Qwen4 architecture, Qwen3.8-Flash features a 125B parameter MoE architecture paired with 51B N-gram Embedding. The model integrates GDN + QSA hybrid attention mechanisms, Gated Residual structures, and the Muon optimizer for the first time, achieving comprehensive performance surpassing Qwen3.7-Plus at just 1/9 of the training cost. It natively supports 262K context (expandable to 1M via YaRN), showcasing hardcore capabilities in programming, office tasks, and multimodal scenarios. Now available for full free API calls in the official group—unlock the brand-new open-source foundation of Tongyi Qianwen at zero cost! Experience it for free now: https://chat.b.ai/chat