anyone actually run LongCat-2.0 on real coding tasks or is it all benchmark screenshots
the 1.6T longcat numbers look spicy but every post i see is a leaderboard cap, zero people showing it fix an actual bug in an actual repo. tried pulling it down this weekend and my box just laughed at me, the active param routing helps but you still need serious vram to feed it. so genuine ask - has anyone here pointed it at a real ticket, not humaneval, and had it hold up across a multi file change? or a big context refactor. curious how far the open side has actually closed the gap on the closed models for coding specifically, not eval theater 😅
every LongCat post is benchmark screenshots, you called it. tried the 1.6T on a couple real refactors and it was fine, not magic, kept wanting to rewrite files i didnt ask it to touch. classic open weights energy
ran the 1.6T on a real repo, same, kept rewriting files i didnt touch. open weights tax. benchmarks looked incredible, actual coding was a solid B minus