Qwen3.5 arrives: a 397B open-weight multimodal MoE with a 1M-token context
Alibaba's Qwen team released Qwen3.5-397B-A17B, a mixture-of-experts model with 17B active parameters, a 1M-token context window and native vision-language support, alongside a hosted Qwen3.5-Plus API. Its hybrid design mixes linear attention (Gated DeltaNet) with full attention layers to cut long-context cost. For teams running open models, it provided a multilingual, agent-oriented alternative that is comparatively efficient to serve at long context.
Read the full story at Alibaba Qwen ↗Like, save or discuss on Buzz
