Qwen3.5 open multimodal models add practical local-deployment options

Qwen3.5-2B and Qwen3.5-122B-A10B provide concrete open-weight deployment paths, including vision, roughly 262K context for the small model, vLLM recipes, and hardware guidance. Modified Qwen3.8 variants claim reduced thinking tokens and faster inference, but developer-reported benchmarks, an AIME regression, and repeated coverage do not yet justify reposting.