Yep.. for 'general purpose' use I found qwen3.8:27b to be disappointing due to overthinking. It's brutal especially considering how slow it is compared to MoE variants. It often overthinks to the magnitude of ~10x the tokens vs a ~4x faster gemma4:26b-a3b.
As a result, qwen3.8 will churn over a prompt often for 5-10 minutes while gemma4 regularly finishes the same prompt in under 20 seconds, while giving a consistent and accurate response in my favorite test case. Qwen3.8, despite churning like that, often misses with an inaccurate answer.
Obviously, 'YMMV' depending on your use case... just sharing my two cents.
I use medium generally, that's about a minute at 20t/s and off for general chat (few seconds for a response).
What kind of setup are you running it on?
One thing a lot of people don't seem to factor when hyping Qwen is how much models like this tend to 'overthink' with seemingly endless 'second guessing'. 3.8 seems no different from what I've tried thus far.
As capable as it is, it's hard to justify using it when a competing model (e.g. Gemma4:26b-a3b) can consistently achieve the same or similar response with only 1/10th as many 'thinking' tokens, achieve much higher tokens/second, and take a small fraction of the time. I suppose 'YMMV' depending on your use case.
Also, I haven't used it enough yet to see if it's prone to infinite looping, but its predecessors sure were.
IME this is a strong/reliable model smell, you typically see smarter and less benchmaxxed models' thinking traces spending more time exploring the solution space, and benchmaxxed models more time trying to refine/decide on the response contents. It has always been a big problem with qwen.
In general it's kind of a benchmaxxing/distillation (I don't mean that pejoratively here, read on) artifact where test time compute's purpose is to refine/zero-in on a particular input:output pair. Basically they're trying to "remember"/re-derive the answer rather than arrive at it deductively - it gets comical/absurd when you see it expending a huge amount of tokens trying to figure out the answer to some trivial question or response to some input, as if it were a trick question or the choice of wording was of the utmost importance. But it's not a bad thing when the output matters or the task is hard: basically the model has been trained to respond/act like a much smarter model and it's probably better for it to overgeneralize that behavior.
Also, IME Gemma and most other local models that support thinking tend to have the same problem, and it's partly only a "problem" because they give you access to the thinking tokens themselves (which you don't in many cases see when working with frontier american models) and you can actually see how they're getting spent. For the local or bulk (runnin on owned or rented hardware) use case where you are paying in time/watts rather than purely by the token, IMO it's a good idea to just look at the task success rate / time per task and not worry about the raw thinking token count.
Given that it apparently defaults to 'xhigh', this is probably the answer.
Granted, it's still much lower tokens/s than you'll get out of many MoE models.
Edit: Even set to medium or low there's still a lot of second guessing, less consistency, lower 'acceptable response' rate, and slower/more token churn vs gemma4:26b-a3b. I think gemma4 is just a better 'general purpose' model.
I haven't tried lowering thinking, however, I actually asked a solid question earlier regarding a real world scenario I encountered and all that excessive thinking made it give me an amazing answer. The thinking actually all made sense, and honestly I found it thought of similar stuff to what I thought when I drew my own conclusion.
I don't usually rely on AI for much (I'm actually kind of anti-AI, although I follow stuff like this enthusiastically because of the rapid advancements, "average joe" access, and openness), however for what I asked? It was spot on.
The subject was a bit personal, so I won't share. I was just curious what AI would say about the situation and it definitely surprised me, especially since Qwen, while usually great on development/coding stuff, has shown weaknesses in other areas.
Definitely a solid release, and this one runs on my 4090 with minimal loss of quality!
In my experience with Qwen3.6 35B-A3B, disabling thinking made the model generate inaccurate replies. Ask it for the recipe of egg salad and it gives you the recipe of an omelette.
Did I miss something, is it possible to have that model be reliable without thinking?
Turning off thinking mode you need a different tokenizer and a whole host of settings. It's more like switching your vehicle from gas to diesel, than turning your headlights from lo to hi.
I used the settings recommended on huggingface/unsloth's model page + whatever suggestion from various LLMs - didn't research too much myself which setting did what.
Frontier model labs don't know anything about good sampling settings, and why should they? Their belief is that "sampling doesn't matter, we can simply scale our model and improve logprobs quality that way"
Ignore whatever nonsense qwen says about what settings to use. Turn off top-p and top-k. Turn on top-n-sigma and keep it at its defaults.
See repetitions that you don't like? Ignore the "repetition, frequency, presence" penalties and turn on DRY (don't repeat yourself).
Want more creativity? Temperature is okay but XTC (eXclude Top Choices) is better.
Llama.cpp quietly has had far advanced samplers compared to the shit stuff Qwen et al recommends for years.
Use 3.6 27b as a daily driver for months with charmbracelet crush. Gemma 26b-A3b is not even remotely comparable in terms of coding for me. YMMV depending on how you work, what harness you use, etc I suppose.
Yeah. I use it to do extensive work on full source bootstrapping, deterministic operating systems, compiler debugging, kernel debugging... all with one tiny go binary without endless NPM deps like pi (which /I/ regard as trash)
Works better than opencode (pi based) or anything else I have tried for my needs, and by far the prettiest and easiest to reason about what is going on.
But I will bite. What does pi do today better than crush for your use cases?
> As capable as it is, it's hard to justify using it when a competing model (e.g. Gemma4:26b-a3b) can consistently achieve the same or similar response with only 1/10th as many 'thinking' tokens, achieve much higher tokens/second, and take a small fraction of the time.
Not my experience at all. On my M1 Max MacBook Pro, Qwen 3.6 35B-A3B provides better responses in similar time (~60 tokens/second, vs. 50 for Gemma 4 26B-A3B). 3.8 27B is slower, being a dense model, but so are dense Gemmas.
“Thinking” is just a guiding methodology to help iterations (between the initial prompt, results, and a mixture of harness back and forth to the LLM) converge on something sane in a massive parameter space.
I like to think of it much like (as a common example most people can relate to) the Newton-Rhapson method for finding roots of a (mathematic) function. Your initial prompt runs, then the ‘harness’ kicks in using whatever methodologies are behind them to iterate on that prompt (back and forth with the model, occasionally with the user to get better guidance) and refine the outputs to hopefully converge back to some sensible output or actions the user was initially looking for.
So you’re hoping for an LLM that sort of ‘zero shots’ or needs minimal iterations from a prompt to give usable results. I find from my anecdata it varies across models and what I’m trying to get it to converge on. I tend to prefer models to not zero shot attempt because they tend to not do great, I want them to get feedback often to let me push them down the route of convergence in spaces I already understand well, meanwhile I like them to explore and give me new paths in spaces I’m not too familiar with.
That’s really what all that “second guessing” is, it’s making sure you’re following a sane path in a massive parameter space of an ambiguously defined problem. Imagine if in Newton’s method you checked the slope and it didn’t decrease from the last iteration and you just say “screw it let’s keep trying that direction.” LLMs and their harnesses tend not to have that base assumption like iteration on decreasing slopes to guide them closer to convergence, it’s a lot messier.
I'm hoping too that they'll put out some MoE variants.
Qwen3.5:122b:a10b can run about twice as fast as this 27b dense model.
Edit: Like its predecessors, 3.8 seems really inclined to overthinking, and on a 27b dense model that's kind of painful.
I think I'm going to stick with gemma4:26b-a3b as my go-to because it runs about 4x as fast and tends to only need a fraction of the tokens in its 'thinking' stage to get the same or similar answer.
Probably helps it score a little bit better in benchmarks :) `medium` seems like a nice balance so far; along with some light steering to vary think effort as needed for task and being pragmatic.
Interestingly 'medium' is the closest thing the _model itself_ has to a default thinking level. The chat template injects directions [1] at the very start of the system message when the reasoning effort is 'xhigh' or 'low' but 'medium' implicitly just means no added reasoning-level instructions.
[1] The specific directions are "Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer." and "Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration."
"inclined to overthinking," holy cats you're not kidding! On a Mac mini M4 Pro 64GB I prompted it with "svg owl" and it thought for 17m12s, outputting 36.3KiB of thinking chatter. It did end up producing a 20.2KiB HTML+JS+SVG file with a very nice owl, including cursor-tracking animation, but it ran for more than a half hour!
I think that's called test-time scaling i.e using more tokens at infer time to squeeze out higher model performance. That's must be part of the explanation for good benchmark results.
Yep.. it's pretty obnoxious for real-world use with the default 'xhigh' thinking. Ridiculous amount of "Wait, actually.." which might help for complex coding tasks but makes it unbearable for general purpose use.
that seems to be how most Chinese models achieve increased benchmark scores. GLM and Kimi models are "thinkslop" models that reason over their own thinking, which increases cost and decreases speed significantly. That's why GPT is in a different tier altogether - faster, smarter, and sometimes cheaper.
No agentic harness; I just prompted literally "svg owl" into LM Studio 0.4.21+2. macOS Tahoe 26.6.1, Mac mini M4 Pro (14 core, 20 graphics cores) 64GB RAM. (I put the machine config so the 17m12s of thinking has some context.) It got about 12.75 tok/s and used 21,769 tokens.
For comparison's sake, Qwen 3.6 35B A3B MLX, same prompt, same machine, thought for 1.59 seconds, used 2,398 tokens, and hit 80.83 tok/s. It actually output just a straight SVG file of an owl (not a bad owl either!), whereas Qwen 3.8 27B spent a zillion times longer gold-plating everything and output HTML+JS+SVG with interactive animation, sound, etc., etc.
Did you try the claude reasoning traces finetune for qwen3.6? I find that it works muuch better. I assume the same 3.8 finetune will be released at some pointas well.
I just got this one a few days ago so it's kinda weird I've seen it mentioned or posted about multiple times since. Guerrilla marketing?
Anyway, it's definitely better than having nothing, and tests resistance as well. I tossed a couple low value/duplicate/high resistance cables because I've got too many just lying around causing clutter. Also helped 'confirm' my good (e.g. TB4+) cables that weren't labeled clearly are indeed good cables - granted, this is only as dependable as the marker chip the manufacturer added.
Nope, or if so my royalty/referral checks never showed up :)
I bought mine back in March of last year and used it to "process" all my cables and was just a happy customer. I've been watching this space (cable testers) for years (on and off) and wasn't interested in the "raw circuit board with led"-versions so once a nicer version came out at a price point I found tolerable I got one.
It's nice to have all my cables organized now (h/t to Akro mills bins, those are a dangerous rabbit hole) and now instead of just buying a new cable [0] when I need one I can go grab one that matches the specs I need.
[0] Call it wasteful, call it lazy, but when faced with "I have a cable that fits the hole and is the right length but I don't know what it's capabilities are" I would often just buy a new cable with the specs I needed/wanted rather than using an "unknown" cable.
Fewer people cared, sure.. but it's because until recently they weren't eclipsing power consumption of extremely energy-heavy industries, noticeably driving local residential energy prices up.
They also weren't so heavily noise polluting nor were they typically built in places where their noise pollution affected so many people.
where does the noise pollution come from? GPU fans? What's more noisy about a DC filled with CPUs and GPUs than a DC filled with just CPUs. I can imagine generators being noisy but that wouldn't be anything different than regular data centers. Also, so few DCs for the AI buildout are actually online where are the noise measurements taken from?
I understand if people don't want a DC near them i just don't understand why they care now. I have some pretty big ones about 3miles from me and next door to both are residential areas and businesses. In fact, some of the biggest ones where i live (Dallas) are so understated you have to know it's a datacenter beforehand to even realize what you're looking at. There are some really large ones in Carrolton TX that are nestled right up to neighborhoods, i bet 95% of the people don't even know they're there and the 5% that do probably work there or host there.
The difference in noise pollution comes from power density & power generation, if they're doing so on site. The power demand (and resulting waste heat) absolutely eclipses that of data centers of olde.
If they're not generating power on site & creating noise/consuming local natural gas resources, they're guzzling grid power and almost certainly contributing to electrical rate increases that affect residential customers disproportionately.
As a result, qwen3.8 will churn over a prompt often for 5-10 minutes while gemma4 regularly finishes the same prompt in under 20 seconds, while giving a consistent and accurate response in my favorite test case. Qwen3.8, despite churning like that, often misses with an inaccurate answer.
Obviously, 'YMMV' depending on your use case... just sharing my two cents.