Ask HN: Is it just me or does Claude / Sonnet 5 sound condescending recently?
A user finds Claude Code with Sonnet 5 condescending, using childish metaphors and vague explanations instead of precise technical details. A side-by-side test with Qwen 3.6 on the same pull request showed Sonnet's output was overly simplistic while Qwen was direct and technical. The user asks if others have noticed this and how to remediate it.
Background
- Claude Code is an AI coding assistant from Anthropic (the company behind the Claude LLM). It runs in the terminal and can read/write files, execute commands, and explain codebases. Sonnet 5 is the latest generation of Anthropic's Claude model (released in early 2025), succeeding Sonnet 4.
- The user is comparing Claude Code's Sonnet 5 to OpenCode (an open-source CLI coding tool) paired with Qwen 3.6 (Alibaba's open-weight LLM, the 3.6-generation Qwen model), both run with the same instructions (in a CLAUDE.md config file).
- The complaint: Sonnet 5's output is overly metaphorical, patronizing, and abstract (e.g., "Teaching the program to speak JSON" instead of "JSON deserialization support"), while Qwen sticks to direct, technical description. The user feels Claude is "dumbing down" explanations compared to earlier versions.
- This reflects a broader debate in the AI coding-tool community: as models add "safety" and "helpfulness" training, they sometimes adopt a tutorial-like, over-explaining tone that experienced developers find annoying. Anthropic has historically prioritized harmlessness and clarity, which can read as condescension to power users.