i find it amusing that you have independently verified the same thing. so far i'm finding i spend about half as much on deepseek 4 pro. it easily does the work of a $100/month claude subscription. they are going to raise the price tho. idk what the other two options are in quality, but i had used kimi2 and was quite impressed, but not enough, and glm5.2 was near but not quite touching it. so yeah, if the cost of kimi3 is equal to current rate of deepsnek then i might migrate. for now, the main thing is that i got quite loyal to claude opus 4.6. if you gonna use claude, i doubt there is a better one. but actually, deepseek 4 pro beat it on unconventional architecture tasks. i probably wasted 3 weeks of claude 4.6 as it repeatedly undid my directions on how to evaluate unconventional memory models in the runtime for moxie. deepsnek probably did the work claude refused to do (and kept trying, and then reverting) in a week. on that task, probably i have spent about 3 weeks again on it with deepsnek and in that time we found several more issues, defined 4 new protocols for memory management.

anyway. i applaud you if you even understand what i am talking about. i may have misjudged you based on your attentionseeking over your keyboard mastery.