Please tell me Qwen3.8-Flash-Next can be made to be better than this

I was quite excited with how well QFN did with some coding and adjacent tasks (like evaluating OSS repos, downloading and running models etc.). Using it for more general research is starting to show that it isnt actually very intelligent and gets lost in its own maze seemingly. Here's a recent example to a simple question: "Have humans always slept for a huge amount of time at once, like eight hours at once, or been multiple sleep sessions through the day from like the early modern civilization?" The verbosity and dense way of writing is too much (feels like a lot of Opus 5 distillation) Setup