Spent the weekend doing some intense research with @NousResearch Hermes and @XiaomiMiMo. Initially, I had GPT 5.4 draft the framework, then I edited the content and let Mimo V2 Pro do a round of review, tweaking the content and polishing the fluency of the Chinese expressions. After a while, I ended up skipping GPT and just using Mimo directly. 1. A big issue with GPT is that it tends to repeat the same point in different ways and keeps referencing the same example across the context (from what it finds online). Mimo points this out, which is super important when writing long-form research. GPT really loves treating what it finds online as gold and then awkwardly inserting it where it doesn’t belong. 2. Even though I specifically told GPT in the memory settings not to give me useless fluff, its output still contains a lot of irrelevant, overly long, and unhelpful content, which makes it exhausting to read. Mimo, on the other hand, doesn’t have this problem at all. It just says what needs to be said, nothing more, nothing less.