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

Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

qwen.ai

Post reply on HN
Print view View original

391–400 of 400 posts

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#391

Post by DeathArrow »

mikenew wrote: post #261
alex7o wrote: post #71

Ok I find it funny that people compare models and are like, opus 4.7 is SOTA and is much better etc, but I have used glm 5.1 (I assume this comes form them training on both opus and codex) for things opus couldn't do and have seen it make better code, haven't tried the qwen max series but I have seen the local 122b model do smarter more correct things based on docs than opus so yes benchmarks are one thing but realit…

GLM 5.1 was the model that made me feel like the Chinese models had truly caught up. I cancelled my Claude Max subscription and genuinely have not missed it at all. Some people seem to agree and some don't, but I think that indicates we're just down to your specific domain and usage patterns rather than the SOTA models being objectively better like they clearly used to be.

>GLM 5.1 was the model that made me feel like the Chinese models had truly caught up. I cancelled my Claude Max subscription and genuinely have not missed it at all.

GLM 5.1 is pretty good but there are some "buts".

They hiked the prices 2 times this year. I subscribed to the pro coding plan just before the last hike. At the start of the year, they had only 5 hours quota and no weekly quota. And I hit the weekly quota hard. I can't upgrade the subscription to get a higher weekly quota because they jacked up the prices a lot recently.

My $30 subscription costs now $72. Previously was $15. Max was $49,then $80 and now $160.

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#392

Post by daemonologist »

greyskull wrote: post #305
daemonologist wrote: post #296

Earlier quoted context omitted.

First of all nothing you can run locally, on that machine anyways, is going to compare with Opus. (Or even recent Sonnet tbh - some small models benchmark better but fall off a bit in the real world.) This will get you close to like ~Sonnet 4 though: Grab a recent win-vulkan-x64 build of llama.cpp here: https://github.com/ggml-org/llama.cpp/releases - llama.cpp is the engine used by Ollama and common wisdom is to jus…

Thank you for all this, I'll give it a shot. Out of curiosity, are there any resources that sort of spell this out already? i.e., not requiring a comment like this to navigate. > nothing you can run locally, on that machine anyways, is going to compare with Opus Definitely not expecting that. Just wanted to find a setup that individuals were content with using a coding harness and a model that is usable locally. What…

Not that I'm aware of. It's kind of like building a PC or a bicycle - you're putting mostly-standardized parts together rather than starting from first principles, but there are so many permutations that you can either use a single known-good configuration or immerse yourself in forums and tinker until you can fit things together yourself. Plus both the inference engines and models are of course moving really fast.

I use Opus 4.7 in Claude Code lol, plus Zed (as a text editor, not a harness). Open-weights models that I can run are for me not useful for multi-turn ("agentic") tasks. I do use Qwen 3.6 for one-off tasks like "write a function to pretty-print this weird data structure" or "explain this config file," and Gemma 4 26B for non-coding tasks like "create a timestamped table of contents from this podcast transcript."

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#393

Post by devilsdata »

renewiltord wrote: post #373
devilsdata wrote: post #251

Earlier quoted context omitted.

I'm Aussie. Please explain to me; why should I care whether Chinese SOEs or the US tech companies are winning? Neither have my best interests at heart.

You will find out when ANZUS ends.

I'm still waiting on those submarines, mate

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#394

Post by syntaxing »

piotraleksander wrote: post #338
freely0085 wrote: post #271

Earlier quoted context omitted.

Composer 2 is just Kimi 2.5, it's not their own model.

it's such a misinformed statement, as kimi2.5 was used as a base model for composer 2 and then heavily RLed

What does heavy RL even mean…similar to how the CEO of cursor said how much better the perplexity got when it’s a terrible metric for model fine tune performance? Let’s be real here, it’s Kimi 2.5 fine tuned for Cursor. There’s nothing wrong with that but they tried to hide it and it’s some work they put in but nothing close to training a model of their own.

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#395

Post by martinald »

greyskull wrote: post #285
martinald wrote: post #281

Earlier quoted context omitted.

Try using a MoE model (like Gemma 4 26b-a4b or qwen3.6 35b-a3b) and offload the inference to CPU. If you have enough system RAM (32GB is a bit tight tbh depending on other apps) then this works really well. You may be able to offload some layers to GPU as well though I've had issues with this in MoE models and llama.cpp. You can keep the KV cache on GPU which means it's pretty damn fast and you should be able to hold…

Thanks! These things you're mentioning like "You may be able to offload some layers to GPU...", "You can keep the KV cache on GPU..." configured as part of the llama.cpp? I wouldn't know what to prompt with or how to evaluate "correctness" (outside of literally feeding your comment into claude and seeing what happens). Aside: what is your tooling setup? Which harness you're using (if any), what's running the inferenc…

Yes fair enough, but try feeding my comment in :). It should be enough for it to go on. Then ask it to explain the concepts I mentioned and ask it to suggest follow-up questions for you to learn more about llama.cpp/local inference!

I've had best results with opencode. Running locally w/ 64GB RAM and Radeon 9070XT (16GB). NVidia should be easier (CUDA), I'm on Linux full time now but used to use WSL2 all the time and had all this working in it.

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#396

Post by make3 »

easygenes wrote: post #315
fwipsy wrote: post #268

Earlier quoted context omitted.

Perhaps not even necessarily subjective, just performance is highly task-dependent and even variable within tasks. People get objectively different experiences, and assume one or another is better, but it's basically random.

Unless you're looking at something like a pass@100 benchmark, the benchmarks are confounded heavily by a likelihood of a "golden path" retrieval within their capabilities. This is on top of uncertainties like how well your task within a domain maps to the relevant test sets, as well as factors like context fullness and context complexity (heavy list of relevant complex instructions can weigh on capabilities in differ…

The pass@100 is such a weird critique angle that is surprisingly mainstream; guess what, no one cares if the correct answer is in the top 100, it needs to be the top 1. A model with a better answer in the top 1 is a better model, full stop.

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#397

Post by make3 »

jxmesth wrote: post #194
embedding-shape wrote: post #192

Earlier quoted context omitted.

> I've tried using qwen and deepseek but they can't even output documents What agent harness did you use? Usually, "write_file", "shell_exec" or similar is two of the first tools you add to an agent harness, after read_file/list_files. If it doesn't have those tools, unsure if you could even call it a agent harness in the first place.

Sorry for the confusion, I was actually talking about their Web based chat. Since most of my work is governance and docs, I just use their Web chats and they just refuse to output proper documents like Claude or Chatgpt do.

there's things like Open Web UI that allow you to easily get a chat UI from an open source model

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#398

Post by make3 »

blurbleblurble wrote: post #321
alex7o wrote: post #71

Ok I find it funny that people compare models and are like, opus 4.7 is SOTA and is much better etc, but I have used glm 5.1 (I assume this comes form them training on both opus and codex) for things opus couldn't do and have seen it make better code, haven't tried the qwen max series but I have seen the local 122b model do smarter more correct things based on docs than opus so yes benchmarks are one thing but realit…

Opus 4.6 was incredible but Opus 4.7 is genuinely frustrating to me so far. It's really sharp but can be so lazy. It's constantly telling me that we should save this for tomorrow, that it's time for bed (in the middle of the day), and very often quite sloppy and bold in its action. These adjustments are getting old. The next crop of open models seems ready to practically replace the big ones as sharp orchestrator age…

I had to write multiple times in my prompt that it's not the model's role to change the subject or end the conversation at all.

I think that they do that to dodge conversations about controversial subjects without full-on refusing to answer. They'll give you an ok answer then tell you to go to get the walk you were talking about.

I also feel like maybe they think people are still ready to pay a lot if they feel like they're getting a lot of "high value stuff" even if the low value stuff the model refuses to do, so they basically try to stop you from doing low value stuff on Opus. I suspect that Sonnet or Haiku never tells you to go take a hike.

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#399

Post by edg5000 »

goodmythical wrote: post #389
edg5000 wrote: post #377

Earlier quoted context omitted.

Very interesting, thanks for sharing EDIT: It's completely different though. This is more of a commodities market/auction/inference broker mechanism it seems.

I mean, the financial incentives are structured a bit different, but it's basically what you're describing, no? It's got projects for data collection, inference, training, etc. It's just that the dollar value of the compute contributed to say training is determined by the value of the token rather than as a straight dollar value. But even that is similar to just renting compute directly via fiat currencies given that…

We're talking about completely different things. I'm talking about creating an LLM in the open, with individual contributors contributing to training sets as well as portions of the training work itself.

Re: Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving

#400

Post by _s_a_m_ »

mikenew wrote: post #261
alex7o wrote: post #71

Ok I find it funny that people compare models and are like, opus 4.7 is SOTA and is much better etc, but I have used glm 5.1 (I assume this comes form them training on both opus and codex) for things opus couldn't do and have seen it make better code, haven't tried the qwen max series but I have seen the local 122b model do smarter more correct things based on docs than opus so yes benchmarks are one thing but realit…

GLM 5.1 was the model that made me feel like the Chinese models had truly caught up. I cancelled my Claude Max subscription and genuinely have not missed it at all. Some people seem to agree and some don't, but I think that indicates we're just down to your specific domain and usage patterns rather than the SOTA models being objectively better like they clearly used to be.

I used GLM 5.1 and it was bad, I have no clue why people claim it is good
Post reply on HN