Qwen 3.6

by Alibaba · Website

Alibaba's April 2026 open-weight generation, released under Apache 2.0 in two sizes: a dense 27B and a sparse 35B-A3B MoE with just 3B active parameters per token. Both are natively multimodal (text, image, and video input) with 262K context extensible to ~1M via YaRN. The 27B's 77.2 on SWE-bench Verified beats Alibaba's own 397B Qwen 3.5 flagship on agentic coding; the smaller Qwen 3.5 sizes (0.8B–9B) remain the current open line below 27B, as 3.6 shipped only these two variants.