Qwen3.5 craves context, skips chit-chat

The real read here is that Qwen3.5 seems optimized for structured, agent-like work rather than empty conversation. That is a feature, not a bug, if you're building tools that need a model to stay on task and use context aggressively.

• –Official Qwen3.5 positioning matches the post’s vibe: native multimodal, agent-capable, and built around a huge 397B/17B active MoE design.
• –The “give it a job” complaint is really a prompt-engineering signal; clear objectives, roles, and long context matter more than friendly small talk.
• –For builders, Qwen3.5 looks strongest as a backend for agents, code workflows, and other tool-using systems that can feed it rich context.
• –The family still looks uneven across sizes, so the flagship’s strengths should not be blindly projected onto the 35B MoE tier.