I have noticed recently that Claude Code often sounds condescending, especially when explaining things. I feel being treated like child, gives me an impression of watching a Sesame Street episode, instead of working with a tool.

Just today I ran Claude Code with Sonnet 5 and, to compare, OpenCode with a local Qwen 3.6, and asked both to explain me a pull request, same prompt, same CLAUDE.md file. Here are some excerpts:

Sonnet title: "Teaching the program to speak JSON"

Qwen title: "JSON deserialization support for AMQP consumers"

Sonnet's description of a larger context: "Rather than one giant program doing everything, the work is split across small applications that talk to each other through RabbitMQ".

Qwen: "This is a Python library that provides shared infrastructure for AMQP data processing applications: applications extend MessageConsumer class, implement process_one_message(), and the framework handles the rest, consuming messages from RabbitMQ, calling the application logic, and publishing results."

Like, from the Sonnet's description alone I wouldn't even know what this code does for real!

And the metaphors by Claude... "From bytes to DataFrames, and back", ugh... Just call it desirialization, come on!

And then later, Claude Code spent a paragraph explaining me how a lookup table is better than an `if .. else` ladder, duh... While Qwen actually focused directly on the nature of the change.

OK, I admit it Sonnet produced a super nice HTML report, Qwen's output is more reminiscent of a man page in HTML format. But... this kind of goes the same way, the niceties just create a impression of being professional.

Did you notice anything like this? Is this new in Sonnet 5, or did I just miss it before? Do you remediate this somehow?