Qwen 3.5 users push back on verbosity

This is really a UX complaint about model defaults, not just a taste issue about writing style.

• –The post frames Qwen 3.5 as capable but inefficient for everyday chat because its answers feel benchmark-shaped instead of user-shaped.
• –Qwen’s own model docs emphasize separate thinking and non-thinking modes, which makes the thread notable because it highlights how wrappers and serving setups can still produce verbose behavior in practice.
• –For AI developers, this is a reminder that inference UX now matters almost as much as raw model quality: concise answers, controllable reasoning, and predictable output length are product features.
• –The comparison to Gemini 2.5 Flash shows why “short by default, detailed on request” is becoming the preferred interaction pattern for fast consumer and developer assistants.