Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Qwen 3.8

twitter.com

Post reply on HN
Print view View original

271–280 of 793 posts

Re: Qwen 3.8

#271

Post by AussieWog93 »

piguin wrote: post #199
mattmanser wrote: post #175

Earlier quoted context omitted.

Comments like this boggle my mind. The model which everyone else raves about and is wildly successful with legions of programmers virtually demanding access while abandoning ChatGPT and Copilot in droves, is rather dumb? Have you considered that it's more likely that you're doing something wrong?

My own experience is that the vast majority of programmers have experience with one model and maybe some short usage of earlier models from a competing choice but want to be using the model with the highest popularity and reputation. I've worked with people who actually had to test multiple choices for their team who didn't understand why they were pressured to select Claude for programmer morale.

I went from cycling between models all the time in Cursor (some would randomly be better at certain tasks than others) to just going pure Opus 4.5 when that came out - it was so far ahead of anything else at the time.

Interestingly with Fable vs GPT-5.6 I think they've lost their lead a bit. I'm finding Fable can't do certain work that 5.6 Sol Ultra can - especially when it comes to webpage design.

Grok 4.5 was fast but made mistakes that GPT/Fable just don't.

I'm curious to try Kimi.

Re: Qwen 3.8

#273

Post by wren6991 »

inkysigma wrote: post #69
storus wrote: post #54

Earlier quoted context omitted.

I would rather see them releasing 3.7-27B, 3.7-122B or their 3.8 versions. Qwen/QwQ were always about the best available local inference at home.

I know this is a bit cliche but I wonder how much headroom there is in the lower parameter count range. Is there any good reason to believe there is a lot of headroom or there is not? I suppose I'm just wondering if this wave of nearly Fable class models will be runnable on ~$10k worth of hardware at reasonable speeds in the near future.

> Is there any good reason to believe there is a lot of headroom or there is not?

It's hard to answer quantitatively, but for example Qwen3.5 -> 3.6 was a significant step in capability, arising from continued post-training of the same models. If we were at the end of low-parameter-count scaling then that would be a surprising datapoint.

Re: Qwen 3.8

#274

Post by taosx »

gchamonlive wrote: post #245
chewz wrote: post #111

Earlier quoted context omitted.

From my experience Qwen-3.7-Max is above the Opus level but delivers results much faster. Slightly worse then Fable. Way ahead of Deepseek 4 Pro (in speed and overall comprehension) - which is a workhorse on its own. I am using them all with Claude Code mostly. Qwen-3.7-Plus is quite OK, good for subagent use. Way better then Sonnet. Qwen-3.8-Max-Preview seems working just fine for me at the moment - I am playing wit…

It's useless to talk about models and harnesses without context and method. Depending on how you use the model and what the model is used for, experience may vary drastically. Also, different models with different harnesses require different approaches. I've been using https://gitlab.com/gabriel.chamon/orisun which is my own simplified methodology, for coding web apps in python and elixir and have been very successfu…

I'm not sure about "useless" but from my experience agentic coding leads to death by a thousand cuts for all projects I've seen so far. Small decisions missed in a codebase that leads to degradation in correctness, reliability and performance. At some point it only takes one engineer to be careless, others skipping PR because they are AI generated...

Re: Qwen 3.8

#275

Post by snake_doc »

adrian_b wrote: post #4

I assume that this announcement has been prompted by that of Moonshot AI, which has just announced a 2.8T parameter open-weights LLM, Kimi K3, to be published on Huggingface by 27 July. Now the response of Alibaba is that they will also publish soon a big open weights LLM, the 2.4T parameter Qwen 3.8. I wonder if Alibaba has always planned to make this big LLM open weights, or they have chosen to do this now, to bett…

Not directly, relevant, but Alibaba (maker of Qwen) actually owns about ~20-30% of MoonshotAI (the maker of Kimi K3).

Re: Qwen 3.8

#276

Post by pixelpoet »

andy99 wrote: post #192
inkysigma wrote: post #69

Earlier quoted context omitted.

I know this is a bit cliche but I wonder how much headroom there is in the lower parameter count range. Is there any good reason to believe there is a lot of headroom or there is not? I suppose I'm just wondering if this wave of nearly Fable class models will be runnable on ~$10k worth of hardware at reasonable speeds in the near future.

Qwen 3.5 to 3.6 was a big jump for the same size, e.g. 29 to 32 on artificial analysis intelligence for the 35BA3B models. Although I don’t think anyone has released a better model of that size since. I would love to see something like a 90B A6B model that is optimized for 128GB machines e.g. strix halo, I haven’t seen anything really targeting the combination of RAM and compute these machines have, but I’m biased be…

Yes, yes, yes! I'm absolutely ready and waiting with dual Strix Halo machines here and really want something approaching Opus at home. Speed is secondary concern for now, that would absolutely change the world.

Qwen 3.6 27b 8b quant 16b kv cache is already pretty good on the Strix.

Re: Qwen 3.8

#277

Post by zobzu »

RazorBucksICO wrote: post #212
chewz wrote: post #125

Earlier quoted context omitted.

> Anthropic should not have bugged their knowledge distillation attacks. It is like one of Pizzaro's men crying that someone have stolen his precious golden dublons As Lenin have said - "Loot the looters" (Russian: Грабь награбленное)

Appealing to the Belsheviks for moral authority is, well I will just say an interesting approach. I do not have that much sympathy for Anthropic, but I do not have much sympathy for publishing companies either whose rights to a revenue stream they violated either. Are Chinese AI companies the Robin Hood in this story? Would they be so magnanimous if they had the upper hand? I don’t think so.

how many hn posts do you believe arent propaganda these days?

its billions, trillions were talking about.

imo hn should display posters origin, such as country, bon, datacenter registered ips, and the discourse will change dramatically.

Re: Qwen 3.8

#278

Post by gchamonlive »

taosx wrote: post #274
gchamonlive wrote: post #245

Earlier quoted context omitted.

It's useless to talk about models and harnesses without context and method. Depending on how you use the model and what the model is used for, experience may vary drastically. Also, different models with different harnesses require different approaches. I've been using https://gitlab.com/gabriel.chamon/orisun which is my own simplified methodology, for coding web apps in python and elixir and have been very successfu…

I'm not sure about "useless" but from my experience agentic coding leads to death by a thousand cuts for all projects I've seen so far. Small decisions missed in a codebase that leads to degradation in correctness, reliability and performance. At some point it only takes one engineer to be careless, others skipping PR because they are AI generated...

All of which you had with only humans in the loop. Catalogue problems so they become technical debt and tackle them periodically. Seems to me like this is less of an AI problem and more of bad management.

Re: Qwen 3.8

#279

Post by apitman »

pixelpoet wrote: post #276
andy99 wrote: post #192

Earlier quoted context omitted.

Qwen 3.5 to 3.6 was a big jump for the same size, e.g. 29 to 32 on artificial analysis intelligence for the 35BA3B models. Although I don’t think anyone has released a better model of that size since. I would love to see something like a 90B A6B model that is optimized for 128GB machines e.g. strix halo, I haven’t seen anything really targeting the combination of RAM and compute these machines have, but I’m biased be…

Yes, yes, yes! I'm absolutely ready and waiting with dual Strix Halo machines here and really want something approaching Opus at home. Speed is secondary concern for now, that would absolutely change the world. Qwen 3.6 27b 8b quant 16b kv cache is already pretty good on the Strix.

What kind of tokens per second do you get on that setup?

Re: Qwen 3.8

#280

Post by NiloCK »

zozbot234 wrote: post #269
baq wrote: post #78

Earlier quoted context omitted.

There’s a Twitter thread making rounds by Dean Ball about deceleration in AI development caused by open models and I can’t understand how people don’t see that it’s true: open models dismantle the frontier lab capex spend potential by reducing the training budget to zero in the limit. Tokens from different providers are not fungible, but customers are nevertheless very price sensitive and close enough is good enough,…

> There’s a Twitter thread making rounds by Dean Ball about deceleration in AI development caused by open models and I can’t understand how people don’t see that it’s true: open models dismantle the frontier lab capex spend potential by reducing the training budget to zero in the limit. If you're worried about an AGI arms race between the U.S. and China putting AI Safety at risk, then the fact that inherently less kn…

The logic, whose premises you can take or leave:

Even at the level of, say, Opus 4.5+, open weight models give a quick turnaround to every Joe and Jane on earth having easy access to pretty high quality improvised weapons design, cyber / auto-fraud capabilities, etc.

All the existing models (closed and open) put up decent resistance to participating in activities like this, and especially behind API walls with content monitoring and account bans.

But the published open-weight models can be fine tuned or abliterated into arbitrarily sharp-edged tools. EG, if it's physically feasible to build a nuke in your garage, it may soon be the case that more or less anyone will have competent guidance to do so.

Post reply on HN