Qwen3.6 Open-Weight Models

Open-weight Qwen3.6 family released under Apache 2.0. Two flagships:

Qwen3.6-27B (dense, 27B, 64 layers, hybrid gated DeltaNet + gated attention) achieves flagship-level coding in a dense model, outperforming the much larger Qwen3.5-397B-A17B on coding: SWE-bench Verified 77.2 (vs 76.2), Terminal-Bench 2.0 59.3 (matching Claude Opus 4.5), SkillsBench 48.2 (vs 30.0). Hybrid thinking mode. 262K native context (extensible to 1M).

Qwen3.6-35B-A3B (MoE, 35B total / 3B active) delivers agentic coding power at dramatically lower compute. Both models support 262K context across 201 languages. Part of Alibaba's dual open + closed strategy alongside the proprietary Qwen3.6-Plus and Qwen3.6-Max-Preview.