Qwen 3.8
Alibaba's August 2026 Qwen generation, spanning a 2.4-trillion-parameter flagship down to a 27B model that runs on a laptop.
Definition
Qwen 3.8 is the generation of Alibaba's Qwen models released in August 2026, covering an unusually wide range of sizes. At the top sits Qwen3.8-Max, a mixture-of-experts model with roughly 2.4 trillion total parameters, only a fraction of which are active for any given token. At the practical end is Qwen3.8-27B, a dense model about a hundredth that size, released under a permissive Apache 2.0 license, which accepts text, images, and video and handles around 262,000 tokens of input natively. The 27B version drew the most attention for bringing near-frontier quality to hardware ordinary people own, and it has become a common base that other labs build on — Cloudflare's Clef decision models were adapted from it.