Khanmigo says that it runs on GPT-4.1, which is an extremely old model with relatively low intelligence. I'd suggest that we update it to a current model which is much, much smarter. A good base model would make Khanmigo significantly better at adaptive tutoring and guiding the student gradually to a solution, which it is relatively incompetent at right now in my experience.

For example, the smallest of the new ChatGPT models, GPT-5.6 Luna, is not only way ahead in intelligence and reasoning capability but also cheaper, if that was a consideration. I don't know about Khan Academy's specific AI pricing options but the standard API cost is $1/M input tokens and $6/M output tokens, compared to GPT-4.1's $2/M input and $8/M output.

Therefore, I don't see a reason not to make this update.  If Khanmigo is to be marketed as a top-tier AI learning tool, it should at least use frontier intelligence to stay competitive.

Source: https://developers.openai.com/api/docs/pricing?latest-pricing=standard