> Why was the dominant narrative on cursor coming within spitting distance of opus with a MUCH smaller team and less capital "LOL THEY USED KIMI!!" instead of "wow, open source models + a bit of RLHF training and some clever context management got within spitting distance of the industry giant and way cheaper"? The latter sentiment is a whole lot more damaging for a company eyeing an IPO with existing investors with very deep pockets.

This comment is interesting because you took a narrative that was being pushed and marketed (Cursor was close to Opus) and accepted it as the ground truth.

The dominant narrative I saw around that, at least in my bubbles, was disappointment when they actually tried it and discovered it was not, in fact, close to Opus.

Those were both narratives going around, but one was clearly winning in terms of volume, and that's what I'm speaking to here. The dismissal towards the open source models has always smelled more like marketing campaigns to me than the actual sentiments of any hackers I know. We all want the open source options to close the gap, but the labs are definitely staying ahead.

My own experience was relatively similar, good, but with a notable gap that went beyond cherrypicked benchmarks.