Qwen 3.5 Review: Alibaba's Native Multimodal Agent Family — 397B MoE, 262K Context, Apache 2.0

Qwen 3.5 (February–March 2026) is Alibaba's most architecturally ambitious model family: nine sizes from 0.8B to 397B, hybrid linear+sparse attention (Gated DeltaNet), native multimodal training, 262K context, 201 languages, Apache 2.0. The flagship 397B-A17B delivers 8.6–19x faster decoding than Qwen3-Max at fraction of the cost. We review the architecture, benchmarks, pricing, and whether it earns the 'native multimodal agent' label.