News for hackers. Stuff that matters.
Re:Qwen4-Max by nelaggy on Sunday November 15, 2026 @10:38PM (#50513737)
is the open weights lag widening? these open models don't seem to be improving as quickly as ant/oai
Re:Qwen4-Max by taneq on Sunday November 15, 2026 @10:41PM (#50513749)
Might just be the commercial offerings scaling up parameters so much? Like for like the open models don’t seem to be lagging.
Re:Qwen4-Max by nelaggy on Sunday November 15, 2026 @11:16PM (#50513957)
either way, ant/oai models dominating the intelligence-cost frontier is not a great outcome for those of us who prefer open weights
Re:Qwen4-Max by omnimus on Monday November 16, 2026 @02:57AM (#50515168)
Even smaller open weights models are already great enough for much of "assist human" work.

This forces ant/oai to lower their prices but how much that is subsidized nobody knows. It might very well just be they are in deep trouble, they see their clients leaving, they have to lower their prices.

But let's not forget openai is riding on the edge of bankrupcy, their last investment round apparently gave them few months of runway, they are leaking money. I don't think they know how to turn this around with all the open models being competitive.

I would just enjoy cheaper model costs while they are like this.

Re:Qwen4-Max by x312 on Sunday November 15, 2026 @11:15PM (#50513949)
I can't imagine them keeping up long term given the information asymmetry plus compute restrictions on top of it.

I feel open-weight was able to keep up the past year using efficiency improvements (hybrid attention, improved MTP drafters, improved optimizers) to offset having less compute. But OpenAI/Anthropic started cutting token costs just a few months after these, which to me suggests they copied these improvements. Indeed it would be hard to work in LLM training and not know about DSpark, Muon etc.

So what do you do if closed source uses your learnings, gives no knowledge back, and has 10x more compute and funding on top of it? The situation feels quite dim right now, esp. with the Nvidia thing.

Re:Qwen4-Max by gimlichgu on Sunday November 15, 2026 @11:23PM (#50514013)
open always wins
Re:Qwen4-Max by pell on Sunday November 15, 2026 @11:39PM (#50514111)
Right now open is heavily subsidized by China and I fail to see what would happen if that stopped.
Re:Qwen4-Max by koe123 on Monday November 16, 2026 @02:38AM (#50515061)
You are subsidizing American AI with your pension after the IPO
Re:Qwen4-Max by pell on Monday November 16, 2026 @07:19AM (#50516816)
Obviously American AI is similarly geo politically supported. However the business model so far is at least somewhat clearer. They invest in the cutting edge and want to sell access to it.
Re:Qwen4-Max by jcmfernandes on Monday November 16, 2026 @07:51AM (#50517080)
American AI has a clear business model: selling you inference.

Chinese AI also has a clear business mode: making money from its (i.e., chinese AI) applications.

I'll argue that the latter is closer to the end-goal here than the former.

Re:Qwen4-Max by jrflo on Monday November 16, 2026 @12:09AM (#50514285)
What advantage do they have over closed? You may want open to win, but logically I don’t see how it would
Re:Qwen4-Max by hellcow on Monday November 16, 2026 @12:31AM (#50514378)
Local models already crossed the threshold where they're quite good and useful for many things, even with cheap setups. I walked away really impressed with (non-coding) use of Byteshape's Qwen3.5-9B quant on my 8GB RTX 3070 for instance. The model was fast, helpful, and 100% private.

If we extrapolate over the next several years, prices for VRAM/RAM will start to come down as manufacturing capacity comes online, models will continue to become more token efficient, and intelligence will improve at every quant size. As this happens, more and more people will be running LLMs locally, and there's going to be a point where local can do everything you need to do unless you're literally trying to do breakthrough mathematics research.

Re:Qwen4-Max by forrestthewoods on Monday November 16, 2026 @01:01AM (#50514516)
I don't think "useful for many things" qualifies as "winning". It's about two orders of magnitude shy of winning.

> and there's going to be a point where local models can do everything you need to do unless you're literally trying to do breakthrough mathematics research

Maybe. But competitive markets pay exponentially more for incrementally superior services. Professional athletes are a great example of this. If you're building a product with competitors and those competitors have frontier models and you stick to local then you're probably going to lose both today and in the near-ish future.

If you're bitter lesson pilled then the alpha is entirely in having a better model, more data, and more compute. Your vision is that human ingenuity will provide more alpha. Oof. I wish. But my tastebuds are very very bitter.

Re:Qwen4-Max by gpt5 on Monday November 16, 2026 @01:19AM (#50514613)
Also - the “open” here doesn’t have the advantage of open source, where it can be improved by the community. At best, the community can tweak it a little. For actual improvements you need to wait for the next model which is trained by companies for billion of dollars, and could stop releasing them at any moment.
Re:Qwen4-Max by pyrale on Monday November 16, 2026 @10:40AM (#50518772)
It can, through better quantizations and reinforcement learning.

Just this week, qwen 3.8-27b got two new interesting quantizations, one of which came from community. Llama.cpp also improves continuously, and novel alternatives like Strata pop up.

Re:Qwen4-Max by byzantinegene on Monday November 16, 2026 @01:37AM (#50514718)
> If you're building a product with competitors and those competitors have frontier models and you stick to local then you're probably going to lose both today and in the near-ish future.

i highly doubt this premise. the success of a product has very little to do with the amount of the intelligence used in producing it. you could throw millions of dollars using AI to develop a movie and it would still flop in comparison to an indie-made film.

Re:Qwen4-Max by wolfinterval on Monday November 16, 2026 @07:27AM (#50516878)
I think the premise is confusing what the amount even is in this context.

I want to use "more intelligence" but not in terms of single higher IQ.

I want to use more intelligence in terms of a sum quantity of knowledge work being performed. A team of 10 instead of a team of 5.

This is when a local model will shine. A fixed cost and I don't have to worry about how much I am using it. I already can't tell any difference at all between Opus and Fable. It won't be that long that I won't be able to tell the difference between a local model and Opus.

There will be an infinite demand for "more" intelligence but I think many people are making a category error on the dimension in question.

Re:Qwen4-Max by avadodin on Monday November 16, 2026 @04:13AM (#50515579)
you don’t even understand the lesson, oofu.

the lesson is that training a model on all input ever output will never be as good as training a model on arbitrary outputtable input.

Re:Qwen4-Max by EddieRingle on Monday November 16, 2026 @01:41AM (#50514734)
I can run an open model on my own hardware, in my own home (which has solar panels on the roof, so electricity usage is mostly a non-issue). Or, if needed, I can rent compute from somewhere like you'd do for any other workload.

Once I have a model locally, I know it isn't going to change on me. Compare that to using an API from Anthropic or OpenAI (and yes, the only thing you can say is you're using an API, not one model), they can discontinue a model you're relying on (as every model could have peculiarities you've designed around) or modulate their speed/quality day-to-day to save on compute costs.

And trying to claim that buying your own hardware doesn't pay off compared to API or subscription costs doesn't hold much water either, at least not for me. For starters, I already had the GPUs and machines I started running models with, whether it was for gaming or for other reasons. And in the case of the new GPUs I bought recently (2x AMD Radeon Pro R9700; no you do not need CUDA and you haven't for a long time now), I can hold a reasonable expectation that they will hold at least some of their value years from now, compared to paying for API/subscription usage for years in which case you're left with nothing.

So, if you have a local model that is "good enough" (and they already exist for many use cases), it's just kind of the obvious choice to make, IMO. And in just one person doing it as a hobby while I'm still looking for a job 16+ months after a layoff. A small business and certainly a large enterprise has even less reason to open themselves up to all the downsides that come with using Anthropic or OpenAI.

Re:Qwen4-Max by indy on Monday November 16, 2026 @06:12AM (#50516290)
They help to prevent the "we've always been at war with Eurasia" problem that would occur if we depended on a couple of 'approved' models
Re:Qwen4-Max by bluebands on Monday November 16, 2026 @01:02PM (#50520680)
I would say AI is more like hardware than software in economics and open hardware rarely wins

why more like hardware? takes very long time to develop, expensive to manufacture (run), valuable.

Re:Qwen4-Max by koe123 on Sunday November 15, 2026 @11:28PM (#50514049)
Show us your stock holdings
Re:Qwen4-Max by alfiedotwtf on Monday November 16, 2026 @12:48AM (#50514462)
It all depends on if continual intelligence with each version is logarithmic or exponential..

.. if exponential, they’ll never catch up but if logarithmic, closed models will plateau and so catching up is inevitable.

My money is on a plateau

Re:Qwen4-Max by bluebands on Monday November 16, 2026 @01:03PM (#50520685)
there is not much evidence there is a plateau

I hope it will plateau

Re:Qwen4-Max by baoooooooooooo on Sunday November 15, 2026 @11:22PM (#50514006)
I think so but it may be a very short term effect, I think their pace has roughly matched the pre-speed up pace of the frontier labs (the 1.7-2x speed up noted on by METR), and I’d guess we see a similar speed up from the Chinese labs once their internal models hit an equivalent level enabling the same speed up.

Following that logic the gap widens a little bit here but then will steady again. Before this there has been a general catch-up trend too which probably also persists leaving us in a slow closing of the gap as the net effect?

Re:Qwen4-Max by gpt5 on Sunday November 15, 2026 @11:31PM (#50514065)
If the speed up is the reason, then they will never catch up (with traditional means). Because it means RSI predictions are real, and we will see an explosive growth in capabilities (far beyond exponential).
Re:Qwen4-Max by orangeboats on Monday November 16, 2026 @01:48AM (#50514769)
>If the speed up is the reason, then they will never catch up

This doesn't take other factors into account. Compute capability, infrastructure capability (the chips can't work without electricity!) etc.

Those factors translate into a ceiling for this hypothetical exponential speedup.

Re:Qwen4-Max by gpt5 on Monday November 16, 2026 @02:44AM (#50515091)
If compute is the bottleneck we’ll encounter in the coming year, then you would expect the gap to increase even more. The US and its companies has three times as much AI compute as the rest of the world combined.
Re:Qwen4-Max by orangeboats on Monday November 16, 2026 @03:41AM (#50515409)
That is if you expect China and US to maintain the same compute gap over time. I wouldn't be so optimistic if I were you.
Re:Qwen4-Max by gpt5 on Monday November 16, 2026 @06:44AM (#50516513)
I’m actually expecting it to expand. There is so much new production being built right now by every company due to the shortage, and China doesn’t have a single EUV production, and only expect a prototype one in two years.
Re:Qwen4-Max by gimlichgu on Sunday November 15, 2026 @11:23PM (#50514017)
ant/oai have stalled in the past 6 mos,
Re:Qwen4-Max by manmal on Monday November 16, 2026 @12:31AM (#50514380)
Do you use them? Opus 5.5 is 2 months old and was a big leap forward. Sol is 4 months old, and was at least a 30% improvement.
Re:Qwen4-Max by raincole on Sunday November 15, 2026 @11:42PM (#50514124)
Yeah, it turns out that the distillation theory has legs.
Re:Qwen4-Max by rbehrends on Sunday November 15, 2026 @11:52PM (#50514177)
No idea, but this is a rapidly changing field. Opus 4.5 was released a little less than a year ago and it feels shockingly limited compared to even current open flash models. (I literally just tested it the other day.)

So, with new models coming out on a literally weekly basis, I think trying to figure out a future trajectory is just going to be a wild guess. For all we know, Anthropic and OpenAI models may also be plateauing in the near future; or there may be a breakthrough and they may improve beyond our wildest beliefs. I honestly don't know.

That said, I suspect available compute and model sizes that major Western vs. Chinese labs can handle will still keep the gap open at least in the short term.

On the other hand, I am also not sure how much the gap will matter past a certain point. I am specifically wondering at what point people will become interested in smaller models fine-tuned for specific application areas (e.g. cheminformatics) rather than massive generalist models.

Re:Qwen4-Max by Aurornis on Sunday November 15, 2026 @11:53PM (#50514184)
This is Qwen4-Max, which is not open weights.

Their open weights models are the the smaller ones, which aren't released yet.

Re:Qwen4-Max by manmal on Monday November 16, 2026 @12:29AM (#50514371)
Ant and OAI hiding their reasoning traces probably doesn’t help.
Re:Qwen4-Max by flincode on Monday November 16, 2026 @08:09AM (#50517234)
I feel like they are improving what actually matters for LLMs, lower memory usage, compute. While we don't know how much far ahead closed source models are in this, but it could be similar they might lower the price to subsidize or make it seem like its all good, either way tech like these being open means it will help us reach further in the long term.
Re:Qwen4-Max by julianlam on Monday November 16, 2026 @10:45AM (#50518845)
I think looking at the lag between frontier and open-weight is looking at an uphill battle and assuming it's equal footing.

Lots of research goes into open-weights, optimizing cache re-use, storage, inference costs, etc., and all these benefits are easily rolled into closed AI models. Meanwhile any advances made by OpenAI, Anthropic, etc. are kept hidden away to give their models an edge.

It'd already be a miracle if open-weights were gaining on-par with closed-weight models. That they've been closing the gap for most of the past year is wild!

Re:Qwen4-Max by sinuhe69 on Monday November 16, 2026 @11:12AM (#50519222)
[delayed]
Re:Qwen4-Max by mlmonkey on Monday November 16, 2026 @12:28PM (#50520187)
Maybe because the frontier labs have closed the ability to distill their models and clamped down on distillation?