5 ms·
Aghhh, I wished they release a model which outperforms Opus 4.5 in agentic coding in my earlier comments, seems I should wait more. But I am hopeful
by throwaw12 9mo ago
Aghhh, I wished they release a model which outperforms Opus 4.5 in agentic coding in my earlier comments, seems I should wait more. But I am hopeful
- wyldfire 9mo agoBy the time they release something that outperforms Opus 4.5, Opus 5.2 will have been released which will probably be the new state-of-the-art. But these open weight models are tremendously valuable contributions regardless.
- wqaatwt 9mo agoQwen 3 Max wasn’t originally open, or did they realease?
- OGEnthusiast 9mo agoCheck out the GLM models, they are excellent
- khimaros 9mo agoMinimax m2.1 rivals GLM 4.7 and fits in 128GB with 100k context at 3bit quantization.
- lofaszvanitt 9mo agoLike these benchmarks mean anything.
- frankc 9mo agoOne of the ways the chinese companies are keeping up is by training the models on the outputs of the American fronteir models. I'm not saying they don't innovate in other ways, but this is part of how they caught up quickly. However, it pretty much means they are always going to lag.
- aurareturn 9mo agoThey are. There is no way to lead unless China has access to as much compute power.
- jyscao 9mo agoThey likely will lead in compute power in the medium term future, since they’re definitely the country with the highest energy generation capacity at this point. Now they just need to catch up on the hardware front, which I believe they’ve also made significant progress on over the last few years.
- anonzzzies 9mo agoWhat is the progress on that front? People here on HN are usually saying China is very far away from from progress in competitive cpu/gpu space; I cannot really find objective sources I can read; it is either from China saying it is coming or from the west saying its 10+ years behind.
- Onavo 9mo agoDoes the model collapse proof still hold water these days?
- CuriouslyC 9mo agoNot true, for one very simple reason. AI model capabilities are spiky. Chinese models can SFT off American frontier outputs and use them for LLM-as-judge RL as you note, but if they choose to RL on top of that with a different capability than western labs, they'll be better at that thing (while being worse at the things they don't RL on).
- MaxPock 9mo agoIf that's how it is done, we'd have very many models from all manner of countries. I mean ,how difficult is distillation for India , Japan and EU ?
- auspiv 9mo agoThere have been a couple "studies" and comparing various frontier-tier AIs that have led to the conclusion that Chinese models are somewhere around 7-9 months behind US models. Other comment says that Opus will be at 5.2 by the time Qwen matches Opus 4.5. It's accurate, and there is some data to show by how much.
- WarmWash 9mo agoThe Chinese just distill western SOTA models to level up their models, because they are badly compute constrained. If you were pulling someone much weaker than you behind yourself in a race, they would be right on your heels, but also not really a threat. Unless they can figure out a more efficient way to run before you do.
- esafak 9mo agoBut it is a threat when the performance difference is not worth the cost in the customers' eyes.