Qwen3.8-Max: A New Bar for Coding and Cowork • Hacker News Board

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#421

Post by cyanydeez » Mon, Aug 03, 2026, 2:01 PM UTC

tommica wrote: post #145

lcnPylGDnU4H9OF wrote: post #121

Earlier quoted context omitted.

Yeah, a good rule of thumb is that the weights take up ~100% of the size of the model, so 100B bytes (8-bit quant) would be, well, 100GB and a 4-bit quant would be half that.

Oh, that is a useful rule to know! Thanks!

That's not exactly the math. Theres also vram needed for context. I operate several 72-128 GB machines and the larger the context the slower they go.

And the context takes space +kv cache. KV cache drives usefulness as your context grows, it needs to pull the kv cache.

Simplified, the context has to be run on every turn, so the KV cache supplies the processed tokens, so it just needs the new inpute.

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#422

Post by nerdalytics » Mon, Aug 03, 2026, 2:02 PM UTC

I used Qwen3.8 Max Preview for 2 weeks and can't tell if I truly miss Fable.

Qwen doesn't overuse em-dashes, weird sentence structures with bold text: normal text.

In my private GitHub Repo Qwen finished as much tasks as Fable did. Without hitting 5-hour session limits. Qwen's token-per-second performance fluctuated greatly. From 20 tps up to 80 tps.

Qwen failed on some "reading between the lines", but so did Fable.

The only noticeable difference is skill loading. Until "Must use" in a skills front matter, Qwen tends to be very lightly loading skills. Superpowers works, because of the same Must use wording. All my local skills use the same wording now. For installed Skills, it's more or less my responsibility to remind Qwen using a certain skill.

I used Qwen together with Qwen Code. I didn't compare Qwen+Claude Code.

With my Owl code scanner, Claude and Qwen produced similar results. Qwen repeatedly used the same output format, despite no format provided and disabling memory. Fable varied greatly. Both followed the file format output and produced valid files.

With Qwen 3.8 being now a visual understanding model, I don't really miss Fable/Opus/Sonnet for my private projects.

At work we are still exclusively using Anthropic models with Claude Code. No change incoming on that front.

Privately, I'm sold on Qwen and Qwen Token plan. No session limits, many open weights models available via a single API.

Since it's horrifying thinking about what US companies do with my data, it's no worse or better thinking about what Chinese companies do. It's a choice between the Plague and Cholera.

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#423

Post by HarHarVeryFunny » Mon, Aug 03, 2026, 2:04 PM UTC

toshinoriyagi wrote: post #8

They've also announced Qwen3.8-27B being released open-weight next week. Qwen3.6-27B is widely regarded as one of the best local models, especially since nothing else comes close to it, that isn't benchmaxxed, without being significantly larger. If 3.8 truly improves upon it that would be awesome.

There was an interesting interview by MLST with a team doing well on ARC AGI 3 who are using Qwen 3.6 27B, and said that it's actually better at coding than the larger 3.6 35B.

I guess which of the smaller 3.8 models is best for coding will depend on which one they put the training effort into.

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#424

Post by jlkuester7 » Mon, Aug 03, 2026, 2:04 PM UTC

westpfelia wrote: post #248

nozzlegear wrote: post #27

Earlier quoted context omitted.

Qwen3.6-35B is my daily driver for AI, and what convinced me to cancel my Claude subscription back in April. The Qwen3.6 line is easily the best local model I've tried, and I've tried a lot. I've got it diligently grinding away on my laptop right now, reviewing and fixing some bugs in my F# code.

I'm a big ole noob when it comes to local AI. What are you using for a harness? Or platform to interact with it?

I recommend trying pi.dev as your agent harness for local models. In my experience it has been the sweet spot of functionality (which you can and should extend with plugins) vs performance (OpenCode just swamps local models on my hardware).

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#425

Post by cyanydeez » Mon, Aug 03, 2026, 2:05 PM UTC

traceroute66 wrote: post #241

pettijohn wrote: post #37

Earlier quoted context omitted.

35B MoE is certainly a good and fast local model. I find 27B dense to be quite a bit smarter, so I daily drive that. I wish there was a ~100B MoE with maybe 10B active. It would be super smart and fast!

> 35B MoE is certainly a good and fast local model. I find 27B dense to be quite a bit smarter Isn't that just the definition of MoE vs dense ?

Full name is 35B-A3B. 3B Is the token generator thats selected out of the 35b available in the model, which is some layered jazz.

So it can be dumber but its quite capablr.

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#426

Post by als0 » Mon, Aug 03, 2026, 2:07 PM UTC

HarHarVeryFunny wrote: post #423

toshinoriyagi wrote: post #8

They've also announced Qwen3.8-27B being released open-weight next week. Qwen3.6-27B is widely regarded as one of the best local models, especially since nothing else comes close to it, that isn't benchmaxxed, without being significantly larger. If 3.8 truly improves upon it that would be awesome.

There was an interesting interview by MLST with a team doing well on ARC AGI 3 who are using Qwen 3.6 27B, and said that it's actually better at coding than the larger 3.6 35B. I guess which of the smaller 3.8 models is best for coding will depend on which one they put the training effort into.

The larger 3.6 35B is actually a mixture of experts (MOE). This means a small proportion of those B's are actually active. It's fast and suitable for agentic tasks but nowhere near good as the dense 27B model, which has all of its parameters loaded.

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#427

Post by clbrmbr » Mon, Aug 03, 2026, 2:07 PM UTC

me551ah wrote: post #257

This makes me wonder if AI companies even have a MOAT in the first place. All requests to an LLM are idempotent, for every API call you need to send it the entire conversation history so that it can process it. LLMs do not learn or remember anything, which makes it super easy for users to switch LLMs on the fly. Most popular AI frameworks, make this a one-liner change these days. And that makes me wonder if the trill…

Mamba/SSMs could change this picture.

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#428

Post by syntaxing » Mon, Aug 03, 2026, 2:08 PM UTC

dionian wrote: post #59

syntaxing wrote: post #49

Earlier quoted context omitted.

I use it with a strix halo server. 35B runs stupidly fast. 27B is about 700 TPS prefill and 30 TPS token generation. Which interestedly is about what Kimi K3 gives me depending on provider.

what hardware do you use or recommend for this? never heard of it until today.

I have a framework desktop, but depending on your need, DGX spark might be better. The prefill and NVFP4 is a significant advantage. But framework desktop is a better general computer. I expect to be able to use it for years to come. Where as DGX Spark you’re at the mercy of NVIDIA BSP.

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#429

Post by msdz » Mon, Aug 03, 2026, 2:09 PM UTC

gcr wrote: post #345

msdz wrote: post #333

Earlier quoted context omitted.

Not who you responded to, but I’ll give it a try. Imagine our LLM as this one-dimensional matrix (instead of unfathomably many, many more dimensions): 0.728201017 -1.9293839303 2.93020202929 5.82290101728 -0.2844920917 … And then, all of these values are multiplied by some scaling factor, essentially vectors pointing in some direction (in our case, only on a one-dimensional line instead of in some higher-dimensional…

This is oversimplified. The proposed question is whether a model with fewer parameters could achieve performance on one language similar to that of a larger model that’s been trained more broadly, which isn’t straightforward to do.

Oh, I didn’t interpret the above question as asking in that direction; but yeah, that’s of course something I didn’t attempt to answer with my comment.

Although I’d be intrigued in the answer to that small-narrow vs. large-broad model question, too!

Re: Qwen3.8-Max: A New Bar for Coding and Cowork

#430

Post by vehemenz » Mon, Aug 03, 2026, 2:09 PM UTC

badsectoracula wrote: post #222

websap wrote: post #216

Earlier quoted context omitted.

[flagged]

A local model needs 0 investment and 0 commitment, takes literal minutes to get started (especially if you have someone who is into that stuff showing you the ropes) and if you end up disliking the experience of using AI you can just `rm -fr` it and forget the whole thing existed.

Local models on regular hardware aren't really capable of anything. Whatever you're testing is nowhere near a measly $20/mo subscription, so it's of limited use.