It does appear to have training for javascript and PHP, from what I can see, and pretty solid knowledge of wordpress and woocommerce. I would guess it has beginner-friendly knowledge of Python, too?

(Though it is gaslighting me about PHP anonymous functions.)

I would not use it to write code (the MoE 26B writes really good PHP), but it appears to have absolutely good enough knowledge to write implementation plans, and I think that could be useful in a sort of agentic coding tutorial environment.

I test these models with simple things. My favourite mini test is asking an AI to write a "last login" tracker facility for wordpress with a sortable admin column, which is trivial code — only a few lines -- but touches on a reasonably deep bit of the WP API. If you ask it to prompt you with clarifying questions, those questions are quite revealing.

It can write the code. Not tested it but I am sure it works. It's not as elegant.

It is not as good at understanding nuanced instructions as either the 26B or the sparse Qwen 3.6. There are concise things you can say in a prompt to Qwen 3.6 that have it draw logical conclusions that fully impress me.

I am more impressed by it than I expected. I reckon this would be quite useful in a tutorial tool.

(I say this as a sort of qualified cynic; I think much of the AI circus is a farce. But if these things are to ever be useful for teaching without making people dependent on some cloud "intelligence tap", this is progress)