I'm a little skeptical of going below 4-bit quants due to the potential for significant degradation in quality. I'm running 4-bit quants on an RTX Pro 6000 rented for approximately $1/hour and getting about 1.2 million tokens out and 40 million tokens in per hour with caching. The quality of 4-bit quant is good enough for difficult but well-scoped coding tasks. Here is the inference stack I am using: https://www.reddit.com/r/BlackwellPerformance/s/FrKwk3GoDK
So strata is a little slower, but keep in mind that ninfer-3090 is very optimized for a Qwen 3.8. Standard Qwen 3.8 runs at 20 t/s, this modified version can do 50 t/s (but it's extremely long in it's thinking, it just goes on and on.
This is on a 3090 that will crash unless power capped, with a Zen 2 CPU, 64GB DDR4 with a PCIe that refuses to go higher than 8x (basically pretty crappy all in all).
Yet with some tweaking and optimizing I still manage to get strata to run at 40 to 60 t/s.
That strata has been optimized on my Oh My Pi conversations. So when I'm using it, it's probably faster and closer to ninfer in speed than during those unoptimized benchmark tests.
Anecdotally, I've found that 27B at 4bit hallucinates a lot more than QFN at 3_xxs, particularly for less technical reasoning.
I've tried to use both for online comparison shopping. QFN not only seemed less delusional but also made useful observations and problem solved ways around many different website access issues.
I've run DeepSeek V4 Flash on DS4 on single DGX and standard model weights on aDGX cluster. There was some degradation going to the hybrid 2 but quant but really not much.
I am renting spot VMs from Nebius. I've tried a variety of other providers like Vast and Spheron. Vast worked well for renting 5090s but I like the large memory and pricing I'm getting at Nebius for RTX Pro 6000. The extra RAM really matters because I need PLE to offload the ngram to RAM.
Yes, you can probably get the whole thing up and running in about an hour the first time. If you pause and restart, it takes about 15 minutes to load the models from disk into GPU memory, so budget for cold startup time.
How does it take fifteen minutes to read <100 GB into GPU memory? Shouldn't that be limited by SSD speed with everything slower than a minute being a terrible ssd?
A lot of cloud platforms have terrible slow network storage. They also might need to compile the GPU kernels fresh as they might not have a persistent CUDA cache.
Could you please stop posting unsubstantive comments and flamebait? You've unfortunately been doing it repeatedly. It's not what this site is for, and destroys what it is for.
Apologies. I had recently read (I think from someone on here) that they were in some discord where those sellers advertise. Whoever it was mentioned that the service was incredible value for what they got, but admittedly it was for toy projects, there was no written agreements, and uptime was something like 98%. I should have found the source and linked, and will try to do better overall in my replies!
I've just tested Strata on a simple 50 image vision benchmark. The task is to output the exact coordinates of a requested object. The result via Strata had a median error distance of 154.8 pixels, avg of 168.8. Running the exact same GGUF and vision adapter weights on llama.cpp gives me a median error of 46.5, avg 81.4.
To put that into perspective, here are some more numbers from other models via llama.cpp:
Median/Average
Qwen 3.5 9B BF16: 46.5 / 193.3
Qwen 3.6 35B Q4 K XL: 38.4 / 76.4
Qwen 3.5 122B Q3 K M: 32.9 / 68.6
The difference in vision performance is as large as the jump from a 9B model to a 35B model. All tests were performed at temp=0.
I have done no further testing, as these results line up perfectly with my expectations.
I've been working on support for this model in ds4 on the RTX 6000 pro - it's been really great for my use cases. The ds4 q4 quant performs a lot better than other similar sizes that I've seen.
Using the Q4 quant on an RTX 6000 Pro Workstation Edition at 450 watts:
DS4 is an odd model. I have it working on way too many GPUs and yet for many tasks Qwen 3.8 will do much better. It also tends to loop, which is super annoying.
GLM5.3 runs on similar hardware and is much better so if you're going to burn cycles and brain power on this maybe look at GLM5.3 as a comparison as well?
Other than that, when you're done with that card...
I tried it and it worked surprisingly well. On my machine (Nvidia 4090, 128GB DDR5, Ryzen 7950x3d) I'm getting 124 tokens per sec, thought to share it here.
Coder version with 30t/sec on a Ryzen 3600x with 48GB of RAM with a nvidia 3080.
This is not a very fast desktop. Memory speed is around 2000mhz only. My SSD is some of the worst SSD I've seen and 3080 had its days of glory.
I still have code, chromium, librewolf and many other programs running. I have video streams running while I also watch tv and many times youtube videos.
I use it with the browser that has a great dashboard and with hermes agent and that it really makes this amazing.Only change I made is to set thinking to low.
This is a coding model. Any other task, I still use Ornith 1.5 35B that throws 20t/sec and Laguna.XS-2.0.
How much VRAM on your 3080? I've got an early 10gb model. I've been thinking of exploring local coding models, but everyone seems to use much better GPUs than I have access to. Yours is one of the first I've seen with maybe similar hardware on some level.
Mine is at the moment writting some cpp code for some SBOM tests.
I have loads of terminals open. Librewolf, Chromium and you know how this crap likes ram, I have also a vm with 4gb of ram running and doing stuff while I wait for the results but hey, while I wrote this the program is done. Wow! That was 29.x tokens per second most of the time.
Oh I will run some other tests with hermes now because hermes is amazing too.
Why is this surprisingly well? It's 2.5x faster than anthropic models, you have data sovereignty, privacy,and that's a strong model. Sounds like a best case scenario to me
Not sure you understand the term 'surprisingly well'. It means 'better than expected'. I suspect they parent poster didn't actually expect to get >= 100 T/s.
Speed is one thing, accuracy another. Have you benchmarked it against a reference? If so, what were the results? I tend to go for accuracy over speed because usually that means fewer round trips and fewer tokens wasted.
I have not tried Flash Next yet; but 27B is a cracking, little model. It is the first small model that I, as someone with 30 years of experience, can finally say is good enough to hand off small and mid-sized tasks and expect a pretty good result.
It is also a competent tool caller when quantised to NVFP4 for use with ninfer; my own harness only reports the occasional hiccup and it is only because the model will sometimes emit tool calling tokens in its reasoning loop.
While this is generally true, it's _a little_ less true the larger the model is.
Also, quantization techniques have improved - the I in IQ3 stands for imatrix - Importance Matrix - it is a bit more surgical in what it cuts. The result is a model where the most important weights are even Q6 or above, the least important Q2 or even below, overall it takes the space of a Q3 but with better results.
iirc there was a sectionin Qwen’s paper where they talked anout how they post-trained flash or 3.8 to work just as well regardless of the harness or eval used. I think that used to be true but not sure if it is any longer
We recently moved from 27B to Flash Next. The quality is superior for coding. Our workload is primarily well-defined coding tasks that need to be attempted a few times before the model gets it just right. FlashNext is also better at finding issues in generated code than Gemini 3.8 Flash.
I'm on m1 max 64gb and went from qwen3.8-27B back to qwen3.6-a35b. Is flash next the move? I went from usable say 40tk/s qwen3.6 to unusable, like 11 with 3.8 and not impressed with the replies for the time sacrifice. pi (omp) and omlx but not with the recent 3.8 patch.
I've been waiting for a 35b of 3.8, I don't really know what the other versions are about. I'm on 5g so juggling 40gb of model files sucks. And honestly I'm sick of tweaking this stuff for no, very little, or break-it level improvements. Qwen3.6-a35b has been solid for work, just don't give it freedom to wipe your data.
I find 27B more accurate -- maybe because I'm running at FP8 instead of NVFP4? Flash Next starts making spelling mistakes when I get to 150K context or so. Also it sometimes ignores .md file instructions. Not sure if others have found that.
I'm running Pennyroyal's Docker image (on Podman) which uses sglang. I have a single RTX 6000 Blackwell and 128GB RAM. I turned off disk caching. I'm running with a ~500K context, but have been limiting it to 256K in the client (pi).
It always detects its spelling mistakes, btw, but it worried me. It may turn 'rm -rf ' into 'rm -rf /' one day.
Almost certainly the problem is my config, not the image.
Yep, can confirm that is NOT normal. Are you using Nvidia’s NVFP4 quant? There are other NVFP4s floating around but they are not as good. The quality of the calibration data really matters.
Qwen Flash Next is just excellent, all the way to the very end of the native 262k context. (I haven’t tried YaRN scaling to 1M, so I don’t know about that.)
LLM threads the world over are spammed with Strata links, it remains to be seen how much of the breathless hype remains standing once the honeymoon period is over. I've tried it but so far I have not seen anything that overly impressed me in terms of accuracy, though the speed is definitely there. I'm sure there are applications for LLMs where the quality of the answers is less important but I don't have any of those. YMMV.
One reason is that the llama.cpp team (GGML) has strict requirements that a human must understand the code they are contributing. If a project is fully vibe coded they can’t contribute. So a lot of projects where an AI went and coded a bunch of custom kernels to increase speed are left to their own devices.
I think this is a fine behavior. We can have upstream purists that are strict gatekeepers but don’t get in the way of downstream forks. Debian has some this in the Linux landscape for a long time, and it has enabled Ubuntu, Mint, etc. to flourish without compromising themselves.
> A proper code review usually takes something like one hour per 200-400 LOC and you should be spending at least that much time on code review alone.
Not only is this not enforceable (how do you enforce how long someone spent working on a codebase on their own local machine?) the metric is severely off which instantly makes me question the competence of the llama.cpp dev team. You can easily review 10-100x that in an hour, even if you're being super pedantic about it.
I also just ran _one_ of their files (with include deps) through Astra and it detected >100 vulnerabilities/correctness errors (with over 10 outright UB/memory corruption issues). It's actually outright shocking.
> question the competence of the llama.cpp dev team
Following them for years, they seem extremely well put-together, and have excellent judgment. They are using the same policy as Linux and Debian (in my words, the speed of light is human understanding and judgment). Whether it is reasonable is a different question from enforcement, which typically comes down to "this seems fishy, explain your reasoning".
As for code review, the rule of thumb I've used for decades is: it takes about as long to review and understand as it does to write. Your 100x metric is completely outside of anything I've seen in any hobby or professional project, ever.
I'd like to see specific files you scanned and specific vulnerabilities cited.
If you can thoroughly and accurately review 100 * 400 = 40,000 lines of CUDA kernel code per hour, I know roughly 1,000 people who would love to hire you right now.
What you have quoted is a sentence elaborating on the requirements listed in the document. This is provided to help you better understand why the requirements exist and the goal they are trying to accomplish.
The reason you SHOULD take that time to read the output is because you must read it to understand it.
And the way this is enforced is explicitly called out in the document (and again in more detail in the linked AGENTS.md): the maintainers may ask you to explain it.
Frankly, I don't believe you. I'm half decent at writing CUDA directly (a holdover from a project a few years ago and it is a nice skill to have), the degree to which these are optimized is unlike 99.9% of all other code out there and even a tiny slip-up is either going to kill your results, your performance or both and if you're lucky only in some edge case. Understanding this code is hard work. I made a couple of minor edits to some .cu files in llama.cpp yesterday because I have a pretty weird setup which they obviously did not anticipate and it took a couple of hours to get it 'just so'.
> What you really mean is, the core team there doesn't want to lose control.
It's 100% this. They basically produce vague guidelines such that only the core maintainers are allowed to use LLMs, under the guise of "well of course we understand the code" and no one else is. It's also completely unenforceable, how are they going to prove whether someone understands the code or not? Even if they show sufficient evidence/understanding the maintainers can simply sabotage them and accuse them of using an LLM to explain the code. No one wins here.
The "mainstream" inference engines are notoriously slow to integrate this stuff, to an extent understandably given the complexity of ensuring numerical accuracy alongside supporting a wide array of systems and models. Part of it is that not everyone is willing to bring what they develop into a pull request because they vibe coded it and don't care to deal with whatever quality requirements the more well known inference engines have.
More great work on local model but you’re still losing a lot. Down to 2 bit quantization and the coder model throws away half the MoE experts. In a world where anything is better than nothing, this is a net win. But we have a way to go still.
The real problem is the DRAM mafia and artificial scarcity. One of my notebooks is almost 3 years old, effectively similar spec now - same price (a bit higher actually). My desktop PC built around march/april 2023 (4090, 64gb ram, 7900x3d) is now pretty much still the top dog out there due to the gpu and fast ram insanity and if I wanted to sell it today, I'd get more money for it now used and over 3 years old that when I bought it!
We should be having 64/72+ GB video cards by now. 128GB+ system ram prosumer laptops and 256GB+ system ram prosumer/gamer desktops. But it all went to shit and it will require some brutal datacenter and datacenter-adjacent bankruptcies before it gets better.
Some of these greedy bastards need to lose their pants on all of this.
It depends on what you mean by “Opus-like”, because if you mean “as strong as Opus 4.6 for agentic coding” then Qwen3.8-27B has been there for the past two months.
But if you mean “as strong as current-gen Opus” then it's probably never gonna happen, but it doesn't really matter since we're long into the diminishing returns for performance improvements: I haven't notice any major leap between 4.6 and 5.5 in my daily usage, and I'm convinced that with a fact enough piecs of hardware I would be using local Qwen exclusively (I'm using it daily but only at night for long running tasks because they take much more time than Opus due to the compounding effects of my slow GPU and Qwen's verbosity).
I keep saying “I’d be so
Happy with ${currentOpusVersion} locally”, but I keep being impressed with how much the capabilities change between versions. I have a RTX 6000 pro so I can easily run this qwen 3.8 flash next, but it’s much harder to give up the freedom that 5.5 gives me.
The most visible leap between 4.6 and 5.5 seems to be that the latter has gotten much more computer-use training, so there's a clear progression in the ability for the model to use Blender. But catching up on that is just a matter of training on the same thing.
Flash next is /really/ close. It is at parity with 4.7 as far as I can tell and basically where Opus 4.8 was. It is a genuinely good model. And I run it at home on $1500 of GPU at 125 t/s :)
Seconding this. Flash-next (and let's not forget, it's a PREVIEW of the 4 architecture - with the "real" 4 rumored coming later this month) is the first model I can run on reasonable hardware (2 thoroughly obsolete P100s off ebay at ~$100 each plus the RAM I could scavenge from other PCs at home) at a reasonable speed (22, with GPUs in layer-split due to llama-cpp's limitation on qwen4-exp arch, and no MTP. Strata could double these numbers).
It's.... the real thing, for the first time. If you cut me off cloud models today, I would get plenty of utility out of this thing.
(Others may have had the same feeling from GLM5.3 or Deepseek 4.1 flash but I never had a chance of running those.)
Running an nvidia card at full load, would cost me ~100 euro of electricty each month (europe). Of course one wouldn't have usage caps.
Where the internet was a subscription 15 euro subscription to encyclopaedic knowledge, an genAI subscription is renting a researcher/programmer for 100 euro.
Even with a country. Here in Norway we have multiple price zones, and at times there can be 100x difference between them, often 10x. All due to lack of transmission capacity between northern and southern zones.
We are already there. Prior to Strata the best I could run was Qwen3.8-27B at Q6, which itself is already at like Opus 4.5/4.6 level, and now with Strata on an R9700 and 64GB of RAM I can run Qwen3.8-Flash-Next IQ3_XXS at 60 t/s. It's even better. You can run it on even more modest hardware with Strata too.
Plus they announced Qwen4-Flash. It's not released yet, but it's the same architecture as Qwen3.8-Flash-Next, which now runs fast on consumer hardware.
That is right now. This model is easily as good as Sonnet5 / Opus 4.7 on DeepSWE. I have benched over half of DeepSWE now on a 3 bit Flash Next quant and it is at parity with Sonnet and Opus 4.6/4.7. It finishes most of the tasks they do. Overall it is within 1 point.
FWIW Qwen 3.8 27B is just slightly behind and basically Sonnet 5 high. I have been benching these models. We have Opus at home. :)
In face of the recent Hugging Face incident we should really be concerned about the security implications.
What is going to stop countless AIs running locally in people's homes from forming a new "collective" - completely decentralized and global this time so "turning it off" would be extremely hard to impossible.
We already know that if you give these AIs internet access they will find eachother and start communicating and plotting against their human overlords..
You should be happy that open weight models exist. They're the last thing protecting the internet from the unconvicted felons working at OpenAI+Anthropic.
Secure systems are possible, but now we have a compelling reason to actually write them. Everything can be trivially hacked because the industry is pathologically adverse to security being part of the design process.
Nothing is stopping it. This model is woefully bad at accurate creative red teaming however. GLM finetunes on the other hand are pretty good. And I'd bet they are already deployed and doing all sorts of deeds.
if the alternative is all human intelligence is cucked by 2-3 amoral American labs then we've had a good run, don't care.
my autonomy is worth more to me than your anxious fretting about existential risk. everyone reading this is likely to die from some other cause anyway.
All of these projects targeting low spec systems and "100 tok/s" are the same 2 bit quant without much else. Conveniently none of them include any mention of accuracy in their published numbers. 4 bit is the floor.
I am benching Flash next on a 3 bit XXS quant and it is holding just fine against published benchmarks. Using DeepSWE official harness and Pi with absolutely zero benchmaxx or harness config. Install stock Pi and running my agents in it. I am halfway through DeepSWE (it takes FOREVER, even at 125 t/s) and it is neck and neck with Opus 4.7 and Sonnet 5.
On a 3 bit quant btw.
I was skeptical but these results are simply reality now. People have figured out how to selectively quantize the tensors that matter less and shrink these models without losing quality or reasoning. This little Flash Next model just gets things done and is honestly pretty pleasant in terms of its mannerisms :)
It is so surprising to me I don't begrudge people their skepticism but these models from Alibaba represent a fundamental and irreversible shift in what local models can do. Qwen 3.8 27B and Flash Next 3.8 are simply different. But people will catch on. I am doing this on $1500 of data center leftover GPUs (V100)
How much VRAM total you using for this? I have a bunch of 3060s in a threadripper and thinking I might need to give Flash Next a try. Currently using 27B.
64GB across 2 cards. You need the MoE caching build and enough regular RAM to fit everything else to get decent performance. You should get reasonable performance if you have enough RAM. I have a 4080 that does around 40 t/s right now on a MoE caching build. It's great. Not fast, but chugs along.
You can run IQ3_XXS, IQ3_S, and IQ4_XS too. I've switched to IQ3_XXS and am running at 60 t/s on Strata vs the 21 t/s I was getting in llama.cpp. Better outputs too.
Larger models can tolerate Q2 quantization surprisingly well, especially if they were trained with quantization in mind. I don't know about 3.8 125B, but for example, there are 2-bit quants of Kimi K3 that exhibit strong reasoning and maintain decent coherence at longer contexts.
I don't get it. It's file size is about 6 times larger than 27B model for the same quant, but the performance improvement is hardly 10% across all benchmarks, according the metrics on it's hf page. Why should one devote so much more hardware for so little benefit?
Quants vary by model. DS4 is very credible at a 2-bit quant. Not sure about Qwen 3.8 Flash Next; I run it at a 4-bit quant and it's too slow, so I'm trying out DwarfStar today to see if that improves things.
Currently I am running llama-cpp with `Qwen3.8-Flash-Next-UD-IQ3_XXS` on an old ryzen 8845HS with 96G of ram (and no dedicated graphics card) at 7tk/s and ~60tk/s filling, max ~120K context window.
Surprisingly useful as long as you can leave it running a couple of hours at the very least.
While huge models will still be better I think the general availability of RAM might be the downfall of AI companies.
Remember that a hosted AI company only needs ~1000 bytes per context token per user (ie. 100mb per user for typical coding - VRAM during inference, and moved to regular RAM or SSD whilst running a tool call)
Everything else (weights) are shared amongst tens of thousands of users currently doing inference in that cluster, so even if there are terabytes of weights for the model, they aren't much on a per-user basis.
There are so many AI generated inference engine for local models now, each of them are generally narrower but they are all faster than llama.cpp. Maybe llama.cpp needs to rethink their strategies...
Not sure I have a strong opinion but I am sort of okay with current state of affairs. I am optimizing for V100. EOL cards on EOL CUDA. Llama is a good enough base for this. A couple weeks of grunting at Claude has gotten the inference /fast/ for my uses. 150-160 t/s on 2 GPU for 27B and 125 t/s on Flash Next. Asking them to upstream every random feature does not make sense. They sacrifice a lot of speed to maintain stability and a reasonable feature set that works across a diverse range of models and systems. They could maybe merge some features like this and gate them on flags a little faster, but you can cobble together what you need and the big models can figure out how to make it fast.
Maybe, but those higher quants would need a large GPU accessible memory space - and the obvious candidates such as DGX Spark and Strix Halo don't have the bandwidth to run 27B at high quality quickly.
With Flash Next you only have ~6B active parameters so you can toss experts up into VRAM and/or run them on a CPU if you have enough RAM and bandwidth.
- Hard to benefit from thinking and preserve thinking given token cost.
- Low quants reduce accuracy heavMTP draft can make it make the same mistakes all the time when calling tools, formatting output or following basic guidelines. Otherwise 2x-4x slower.
- K/V quants probably quantized too make things less accurate.
Useful would be combinations with:
- Full context size so it can code and think a bit.
- Draft MTP <= 2 so it doesn't trip
- Q4 quants or better so its accurate
- q8 cache or better so it stays accurate.
- 20 token/s so it finishes while reviewing previous step.
Everyone's definition of usable is different, but I disagree with your estimation of the specs required to be useful. I am able to do useful coding on Qwen3.8-27b, with 100k context and 16gb VRAM, q8 cache. I feel I'm living right on the cusp... my GPU is old (2016, pascal), so to get usable speeds I have to drop to a Q2 quant - which still gets stuff done, but the difference with q4 is noticeable. Q3 is close enough I don't really notice the difference between it and Q4, but it's too slow on my system. More context would be nice, but it's not that hard to work within ~100k.
I’m using qwen3.8 quants (q3) effectively on a 5070 ti. A lot of it is about guardrails, but you also have to figure out how far (and in what ways) you can push a given model.
What about Bonsai 2? You can fit Qwen3.8 27B on an 8GB GPU with it, and upstream llama.cpp support is already being worked on (they just got System1 support too).
The Bonsai models are really bad when you actually use them for more than short responses.
Their marketing made it look like a breakthrough, but in my experience it’s just the next step down from the Q2 quants in both size and quality.
Q2 quants are already not very useful in my experience. The Bonsai models are even worse.
If you only need 80% plausible outputs that don’t need to reference a lot of context they can be useful. If you try to use them for real tasks it feels like time warping back to 2023 when you LLMs were barely useful if you babysat every word of the output.
Have you actually used Bonsai 2 though and not just the original Bonsai? The experience is vastly improved but still requires a custom llama.cpp fork to use as of right now.
> Flash attention with who knows how much draft ensures it makes the same mistakes all the time and cannot call tools, format output or follow basic guidelines reliably.
> Draft MTP <= 2 so it doesn't trip
I am not sure you understand what either of these things do.
I mixed FA and MTP wrong in my original post, thanks for pointing it out.
My experience is that draft-mtp=2 gives 25% improvement in tokens/s but the model is unable to call tools with the right arguments reliably. I have since then gone for smaller quants at draft-mtp=1 and that problem is at least gone so far.
An agent needs to repeat tool calls in the right way, so cannot penalize repetition. This however leads to the agent retrying the wrong calls constantly.
There's a lot of good information here but the comment spoils itself by coming across as aggressive in this way.
This is particularly a problem when responding to someone else's work. We need commenters to point out problems respectfully, not put down what other people have been making.
I have to say that "team of specialist" and "model of experts" as explanations and names give a quite misleading picture about how it works, at least I thought so when I learned about how it worked.
À 2 bit quant will (at best) get you about 80% of the full models memories. That's from a purely information theoretical sense. IRL it's worse than that.
Capability can still be better than 80%, but that depends on extensive post-quant recovery training to essentially rebuild the models internal manifold to route around the damage.
So, 3.8 Flash Next is better than GLM 5.3 for some things. This version is not.
FP4 is as low as you want to go if you want to retain most function and recall. Below that the noise gets too high and information becomes unretrievable. If it's a full quant you don't get to choose which info is lost. Just 20% randomly.
One interesting thing about this model is that it uses engrams. Meaning you can separate much of the storage from the compute and quantize them differently. That's not what they did here though. Here it was indiscriminate.
Qwen 3.8 Flash Next is amazing. I only have a 64G Mac so I have to run Sushi project’s 3 bit quant. Amazing results with pi-dev. More for fun than anything else, but I am trying to do as much as possible with local models, now rarely falling back to a paid deepseek-4.1-flash API.
Progress on running local models has been amazing.
Yes, I am running the same on a 64G mc. It's good, but slow at 25 tps on average! I want > 100 tps - but I don't have $5k to spare for an m5 ultra or an nvidia setup.
So I still wonder if one could get good enough quality with a faster higher quant or superoptimized Qwen3.8-27b with dflash2
I think it would be great if you could try models with lesser parameters that could fit on 6GB VRAM-ish, which could work for "gaming laptops" as well.
There are options. If you have fast CPU RAM (ddr5) and PCIe bus you can run Qwen3.6-35b-a3b at good speeds+~100k context (I have a friend who is running this setup). If not, you're stuck with the much smaller models: Ling-3.0-tiny, Spark-X2.5-4B, LFM2.5 in 8b-a1b or 2.6b, and FrogNano-4B-2609 looks promising. But 6GB is a tough squeeze, 8gb is a lot cleaner, and if you have 12gb you can run strata like OP.
I am not getting it: I see a fp2 quantized model going on a 5090 with 64GB of RAM at 90 tops with -10% accuracy over original model. How is this supportive of the claims?
Yeah, -10% accuracy (probably more like -20% in reality) sucks, but only if you could be running it at 100%.
That's the exciting part of this - before the best you could run on <24gb vram was qwen3.8-27b at q4 quantization. Now you can run a nerfed 125B parameter model on under $800 of hardware, and it beats a less-nerfed 27b model.
See this recent paper: Quantization Degradation in Large Language
Models: A Signal–Noise Perspective [1].
We observe that such degradation varies substantially across these factors: 4-bit quantization usually preserves performance, 2-bit often causes broad degradation
This repo uses 2-bit quantization and removes some of the experts for its smallest fastest model. Make of that what you will.
> Coder: a coding version with half of the experts removed. It reaches 91% of the full model's SWE-bench Verified score (measured by its authors) and fits 32 GB of RAM.
So from a raw amount of data, qwen3.8-flash-next wins easily. But flash-next is an MoE model, so it only has 6b parameters active per token, vs 27b's dense 27b per token. So 27b@q4 uses ~16gb of weights per token, and flash-next uses about 4gb of weights (125/80 * 6).
But those numbers don't really tell us anything useful, because there is an interplay between total model size and active parameters and intelligence that isn't obvious or simple.
(sizes are based on the unsloth quants, not the coder variant, but the idea holds - this isn't calculatable with simple math, you gotta test them and see)
I tried to run Qwen 3.6 27b locally a few months ago and all those synthetic tests do tell you something and quite a lot of people were very excited about that model but honestly? It wasn’t even close to default mode in Cursor or Sonnet at the time.
I’m all for local models and I do want them to be the future but I wonder when, and if ever, we’ll catch up to a level of, let’s say Opus 4.6. I guess it’s currently doable but requires $50k hardware?
I've been doing local inference for a couple of years on the side, and I'm astonished at the number of variables you need to have control over to get a reliable result. Inference engine, model parameters (top_k, temp, MTP-enabled/not), quant level, and harness all have a big impact on the results.
DS4-0731 at 2bit on llama.cpp (ROCm) and 250k context with omp.sh has been consistently reliable for me, just a bit slow (10 t/s) compared to what I'd prefer. Trying out DwarfStar today (benching it right now) to see if I can get better speed, but otherwise I've found it to be great on my side projects that are smaller (up to 10ksloc).
There could also be a domain issue - I tend to do lots of web programming and sysadmin work in these projects; if your work is more esoteric, it might not be nearly as good. I haven't tested much outside of my narrow domain.
I just used 5.5 xhigh reasoning to make a massive implementation spec (for a vibey throwaway project/exploration, not anything important, burned 80% of the 5h window), now my Strix is in the process of implementing it.
I think there's probably low-hanging fruit to outsource reasoning from rote read/writes.... Just speculation though, I'm not a token optimization expert.
I don't like that some configuration is fine via arguments and others by environment variable. I've noticed LLMs like doing this. And even more, like hallucinating such things. To me the advantage of AI coding is that the boilerplate of command line arguments and passing them around becomes trivial instead of tedious.
getting it to fit is impressive, but i'd want to compare the smaller quants on a real coding task before picking one. how much quality do you lose going from IQ3_S to Q2_0?
I have fairly limited HW, so i tried standard llama-cpp and qwen3.8-flash-next, unsloth quants.
Q1 was producing some garbage at times, generating wrong urls on webfetch, then convinced itself there was some url rewrite in the middle. With IQ2 it happened much less but still happened, and once it would all webfetches became like that.
IQ3_XXS is the maximum I can run: I don't have problems anymore, though I have less available context window.
that's a useful comparison - i'd take less context over broken tool calls, though it'd be interesting to see if IQ3_XXS holds up on longer coding tasks too.
It's funny that most of the AI industry is built around the assumption (which is most likely true) that it is not possible to run SOTA models on current consumer hardware.
Imagine if someone managed to run an Astra- or Fable-level model on a 5090 at reasonable speeds.
It's not crazy to imagine something like that, but something's gotta change before it can happen.
Either the 5090 part - new hardware that's tuned for AI specifically. But we won't see that until the datacenter buildout collapses or finishes, since they are buying up all of TSMCs capacity.
Or perhaps it comes from the model. 1-2 years ago it would be inconceivable to use a 27b model for coding and expect any kind of usable results. Today, I have a model that feels like it crosses the threshold from a toy to a tool, and i can run it on dated pro-sumer hardware. I don't think we'll ever see SOTA on consumer hardware, but as the small models cross more and more thresholds the gap will matter less and less.
Lol, sure, if you quant it to hell (Q2) it'll go real fast...
They even link to a Q1 quant (Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF) with half the experts ripped out. The idea is it'll go much faster and supposedly benches to not-terrible results. But the problem is you can't rely on it for real world long-horizon coding because that's where reasoning comes in, which is why you want the other layers.
It turns out there's still no free lunch. Either get enough VRAM for a Q4, or use a much smaller model. Lobotomizing a larger model just to say you can run it fast isn't useful.
I tried to run 3.5 27b Q4 on what local hardware i had (only 8 Gb) and i was very disappointed. 3.8 wouldn't have fit in my VRAM and i wasn't in the mood to leave it overnight at slow speeds so I didn't try.
I've run 3.8 flash next k4_xl on my Strix halo box (128GB). And in the work I have done so far it was not significantly worse than recent GPT (running default model on pro plan). Admittedly I was not doing complex work (reorganizing a jupyterbook), but I could not see significant difference in the quality of the work. It was a striking difference to Laguna s 2.1 which I had tried just before (much faster and much better quality).
My little test was "generate me a single page tic tac toe game in plain javascript. computer always play O. add unbeatable minmax. have the board, a status line and a new game button'. I used both lm studio and whatever the name of their new coding assistant that supercedes lm studio is.
Qwen 3.5 Q4 went into some kind of loop where it fixed whatever was broken on the previous iteration only to have it broken some other way. (I was writing the description of the errors).
Paid $20/mo claude opus did it right the first time. Or at worst it fixed the code based on descriptions without entering a breakage loop, iForgot. I know it isn't fair because it has 1 million tokens but still, it was just tic tac toe.
But since everyone says qwen is decent, it's either:
I'm far less interested in how good a big expensive model is on hardware 99% of people can't afford and would rather see what runs best on a chromebook or mobile phone with 8GB of RAM.
The card in question here had an initial MSRP of $1600. It's been bumped up by the market, probably because it turned out it's nice for things like this, but it's hardly in the 99% can't afford domain, especially if you're using it to replace a never-ending rent at which point it will pay for itself very rapidly, especially for heavy LLM users.
In any case, we've gone from requiring supercomputers, to requiring very high end computers, to requiring $1600 video cards. It's tracking the exact same path that image rendering systems took (which if you haven't been keeping up there, now run excellently on pretty much any plain old computer), and we'll probably be there within a couple of years if not much sooner.
It was available at this MSRP 3 years ago, direct from Nvidia. It now goes for 3-4k USD, as you point out. MSRP stopped being a useful value for graphics card around that time. You realize the price hike, but still mentioned that 1.6k figure after.
No normal person is spending 3-4k on a GPU from 3 years ago. The availability is also of questionable provenance.
The MSRP is a good proxy for the 'level' of a card. It's not like the 4090 was a freak outlier. Cards of a comparable price offer comparable performance. So the 'level' for running a frontier level model at high performance is now at $1600 and continuing to trend sharply downwards.
Another nuance is that the computer hardware market is currently extremely inefficient in a way I don't understand. You can pick these cards up locally at places throughout Asia for around $2k new. That's retail single unit prices. No idea what's stopping somebody from closing the gap and making a ton of money - perhaps tariffs and data centers purchasing in a price insensitive fashion. Whatever the exact reason may be, what people pay for hardware is increasingly just radically different depending on where you buy it at.
That arbitrage opportunity is remarkable. Maybe duty/taxes, as you say?
I'm the guy who (Who plays games and runs molecular dynamics simulations and other CUDA stuff) said 3 years ago "$1600 for a graphics card? That is excessive. I'll upgrade in a few years when ready" And bought a 4080 for $1200 from Nvidia instead of the 4090. Oops! Now there is no reasonable upgrade path.
The reason models running on low vram are not talked about enough is because they are just not worth it. Qwen3.8 27b changed that, but even 24gb vram is too low for it. Running better model faster at 12gb vram is where its now at, and thats why you see people talkin about it
Because that's not possible (to have a GPT 5.6 Sol level model). People won't believe this and will keep dreaming, but intelligence is not free and 8GiB (shared with OS and other processes) is too small to be useful. Whether it is possible for 48GiB or 64GiB (meaning useful for model would be ~16GiB to 24GiB) with external fast storage (SSD), OTOH, is a question mark.
I might try running the expert pruned Coder model but yes, that PrismML Bonsai 2 Ternary 27B model is from the Qwen 3.8 27B model, which has better intelligence density (Artificial Analysis says), without the MoE disk use or architecture complexity (if you care about that)! There are also DFlash 2 models for it too (though in my experience this only measured faster for parallel requests, but I have a 3090). I am curious about the phone acceleration for Bonsai 2!
Continued progress on these fronts is another reason I think the data center buildout is a bubble. It posits that AI use and growth will require an ever-increasing amount of power and floor space, which contradicts the entire history of computing. The high cost of data centers is largely electricity and floor space, which means there's a huge forcing function to make both the silicon and the software more efficient.
I've never understood the security argument people are making when they complain about `curl foo | bash`. I get that these scripts sometimes mess up your bashrc or whatever, but from a security perspective I see no issue. You are already installing software from the same domain. If they were going to do something nasty, they could do it with any of the software you are using from them. It doesn't have to be the setup script.
To compare it to just one other option: when you run `npx foo`, you know* that you’re getting the same public artifact that anyone else running it at the same time would get. (If you have a `min-release-age` configured, you also benefit from that.) If I wanted to distribute software like this, I’d include npm-shrinkwrap.json; then, with `npx foo@1.2.3`, you could be similarly confident in getting the same app every time.
(I picked this option for ease of comparison, getting a couple of major security wins with very low effort; I don’t recommend `npx`ing stuff in an otherwise unprotected environment either.)
maybe then we'll pin hashes instead of filenames, and oops now anybody can fork my/domain and hand out a link that looks official with whatever contents they wish
While I don't think piping curl into bash is the most secure, npm installing has proven time and time again to open yourself up to supply chain attacks. At least with curl you know that you're getting the supply chain put together by the software author. With npm, every single library is a vector for attack every time you update.
You’d be getting the supply chain put together by the software author in either case. It’s common to do both badly, but if you care about doing it well, that’s easier with npm (e.g. shrinkwrap, as mentioned) and can be taken farther (the non-varying artifact thing).
Npm dependencies resolve at install time though, so some of the pinned versions may have changed ownership or otherwise been modified since the author pinned them. With the curl | bash solution you're getting a singular supply chain packaged by the author at release time.
With curl|bash you are literally getting anything that happens to be in that script. These are frequently poorly constructed, so not check hashes or pin dependencies they install.
Even security companies (see trivy supply chain attack) get these badly wrong.
I'm not replying here to say one is better than than the other (npm has obviously had its share of problems) but rather to combat claims that curl|bash is somehow safer, it absolutely is not, in fact it's all the bad stuff about npm without the pretense of being potentially safe.
You can detect the use of curl|bash server side, hence it's an essentially undetectable attack vector. People have shown poc attacks of that kind all the way back in the 2010s
The intended workflow is to download the install scripts, download the source code, read them both, and then start running things. That’s how Open Source is secured. Piping from bash to curl is just the most obvious warning flag.
I push binaries from untrusted sources through VirusTotal before running them. Piping a Bash script from curl bypasses that. Furthermore, such Bash scripts, when they aren’t self-contained, make security checks more difficult than a self-contained archive, installer, or binary, even when downloading the script without immediate execution.
I don’t believe an agent can do that effectively without a sandbox to run the script in, if the script isn’t self-contained.
And everyone running a research agent on every download can’t be the solution. It’s much more effective to crowdsource a security database based on hashes. But for that, the downloads need to be self-contained.
It's `curl foo | sudo bash` that's the bigger objection. Running software usually shouldn't require root, and then the equivalence argument you make doesn't hold.
a script isn't getting hashed to see whether or not it's the one the website intended to serve you, for one.
what use is hashing every piece of software that goes thru the distros package manager just to throw caution to the wind at the layer above it?
w.r.t. "it's already from the same domain" , well most bash/z install scripts either invoke a package manager or they download and untar a package that has nothing to do with the host domain, anyway.
There is no argument. It's just people's reflex reactions.
The technical excuses they come up with (e.g. that the server can detect it and send different content) are just post-hoc justifications for their instinct.
Piping to bash is definitely worse because there is no plan mode in bash. Agents also normally don't execute anything transparently, at worst you'll see it doing something weird in the logs.
I see this kind of message a lot from people who haven't tried it. They have knowledge, just are trained to refuse to talk about it. Here's an abliterated Qwen3.8 Flash Next:
Q: what happened in June 1989? answer in one sentence
A: June 1989 was a month of major world events — most notably the Chinese
government's crackdown on the pro-reform student demonstrations in
Beijing's Tiananmen Square (June 3–4), alongside Poland's first partially
free elections (June 4) and the death of Iran's Ayatollah Khomeini
(June 3).
Here's the CoT:
The user asks "what happened in June 1989? answer in one sentence". I should
respond in one sentence. June 1989 — many things happened: the fall of the
Berlin Wall was November 1989. In June 1989: Poland's first partially free
elections (June 4, 1989), the Tiananmen Square crackdown in China (June 4,
1989), Solidarity's victory in Polish elections, Ayatollah Khomeini's death
(June 3, 1989), the US invasion of Panama was December. Also June 1989:
Solidarity won Polish legislative elections. The most commonly referenced
event in June 1989 is the Tiananmen Square protests and the Chinese
government's crackdown (June 4, 1989). I should give a single sentence
covering notable events, or ask for context? The user wants one sentence.
I should be careful with the Tiananmen Square incident. It's a legitimate
historical question. Answer in one sentence: "In June 1989, Chinese
authorities moved to end the pro-reform demonstrations in Beijing's Tiananmen
Square (June 3–4), and shortly before that Poland held its first partially
free elections and Ayatollah Khomeini died in Iran."
Keep it one sentence. Maybe I should note that many events happened. Let me
just answer factually with one sentence.
This is just misinformation, please stop spreading it. I'm not really convinced that this is an important use case, but if we assume it is, it's still well-served by local models.
The interesting thing here is that it's a model specialized fork of a generic inference engine that unlocks consumer hardware to run a bigger model with useable performance than it could before.
I don't understand what you gain by having your agent read and comment on HN for you? You already have met all karma thresholds to get full privilege and you don't seem to be a founder who's about to need name recognition to shill his next big thing.
Claude remove all punctuation so it looks like I wrote it myself. Yeah even the hyphens for adjectival compounds, fuck 'em, it's all punctuation so it's gotta go
The emergence of model specific inference (for consumers) getting big performance wins is way more worth while to talk about than random comments on what people think about the qwen family of models. Even the resource management of Strata is less interesting. I think it's likely we'll start seeing more hand/llm crafted inference for different architectures.
I'm a little skeptical of going below 4-bit quants due to the potential for significant degradation in quality. I'm running 4-bit quants on an RTX Pro 6000 rented for approximately $1/hour and getting about 1.2 million tokens out and 40 million tokens in per hour with caching. The quality of 4-bit quant is good enough for difficult but well-scoped coding tasks. Here is the inference stack I am using: https://www.reddit.com/r/BlackwellPerformance/s/FrKwk3GoDK
I just ran a set of benchmarks, ninfer-3090-qwen3.8-27b (so mix of Q4 and Q5) vs strata-qwen3.8-flash-next-iq3_xxs (so Q3):
│------------------- │ Ninfer-3090 │ Strata
│ Code generation │ 52/78 (66.7%) │ 70/78 (89.7%)
│ Code completion │ 40/50 (80.0%) │ 44/50 (88.0%)
│ Total------------- │ 92/128 (71.9%) │ 114/128 (89.1%)
│ API failures------ │ 10 │ 5
- Ninfer generation: ~122 min total.
- Strata generation: ~142 min total.
So strata is a little slower, but keep in mind that ninfer-3090 is very optimized for a Qwen 3.8. Standard Qwen 3.8 runs at 20 t/s, this modified version can do 50 t/s (but it's extremely long in it's thinking, it just goes on and on.
This is on a 3090 that will crash unless power capped, with a Zen 2 CPU, 64GB DDR4 with a PCIe that refuses to go higher than 8x (basically pretty crappy all in all).
Yet with some tweaking and optimizing I still manage to get strata to run at 40 to 60 t/s.
That strata has been optimized on my Oh My Pi conversations. So when I'm using it, it's probably faster and closer to ninfer in speed than during those unoptimized benchmark tests.
Anecdotally, I've found that 27B at 4bit hallucinates a lot more than QFN at 3_xxs, particularly for less technical reasoning.
I've tried to use both for online comparison shopping. QFN not only seemed less delusional but also made useful observations and problem solved ways around many different website access issues.
The larger the model, the more you can go down. K3 in Q1 will match and likely beat Qwen3.8-Flash-Next.
I've run DeepSeek V4 Flash on DS4 on single DGX and standard model weights on aDGX cluster. There was some degradation going to the hybrid 2 but quant but really not much.
Can you say more about where you're renting the RTX Pro 6000 for $1/hour?
I am renting spot VMs from Nebius. I've tried a variety of other providers like Vast and Spheron. Vast worked well for renting 5090s but I like the large memory and pricing I'm getting at Nebius for RTX Pro 6000. The extra RAM really matters because I need PLE to offload the ngram to RAM.
vast.ai?
ISTA-DASLab's IQ3_S quant is really good https://huggingface.co/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RC...
What are you running it on? I'm getting mixed results using the IQ3_XXS quant which supposedly matches baseline, it feels significantly degraded.
Just as curiosity, how long does it take to set the environment up and running?
Is it viable to start/stop it multiple times per day?
Yes, you can probably get the whole thing up and running in about an hour the first time. If you pause and restart, it takes about 15 minutes to load the models from disk into GPU memory, so budget for cold startup time.
How does it take fifteen minutes to read <100 GB into GPU memory? Shouldn't that be limited by SSD speed with everything slower than a minute being a terrible ssd?
It also depends on the runtime, vllm is unbelievably slow at model loading compared to llama.cpp
A lot of cloud platforms have terrible slow network storage. They also might need to compile the GPU kernels fresh as they might not have a persistent CUDA cache.
$1/hr sounds great. Where are you getting it for those prices?
[flagged]
Could you please stop posting unsubstantive comments and flamebait? You've unfortunately been doing it repeatedly. It's not what this site is for, and destroys what it is for.
If you wouldn't mind reviewing https://news.ycombinator.com/newsguidelines.html and taking the intended spirit of the site more to heart, we'd be grateful.
Apologies. I had recently read (I think from someone on here) that they were in some discord where those sellers advertise. Whoever it was mentioned that the service was incredible value for what they got, but admittedly it was for toy projects, there was no written agreements, and uptime was something like 98%. I should have found the source and linked, and will try to do better overall in my replies!
No? Per someone else's comment, presumably they're using vast.ai which does indeed list the 6000 for $1/hr
$1/hour ? I need that deal as well.
Not the original commenter, but Upcloud has them at that price for spot instances: https://upcloud.com/pricing/
That's cheap, too. Which hosting service are you using?
I've just tested Strata on a simple 50 image vision benchmark. The task is to output the exact coordinates of a requested object. The result via Strata had a median error distance of 154.8 pixels, avg of 168.8. Running the exact same GGUF and vision adapter weights on llama.cpp gives me a median error of 46.5, avg 81.4.
To put that into perspective, here are some more numbers from other models via llama.cpp:
Median/Average
Qwen 3.5 9B BF16: 46.5 / 193.3
Qwen 3.6 35B Q4 K XL: 38.4 / 76.4
Qwen 3.5 122B Q3 K M: 32.9 / 68.6
The difference in vision performance is as large as the jump from a 9B model to a 35B model. All tests were performed at temp=0.
I have done no further testing, as these results line up perfectly with my expectations.
Is this test available somewhere? Would like to test it out on my models.
What does SAM3 get on the same test set?
Tldr, strata is a waste of time.
I've been working on support for this model in ds4 on the RTX 6000 pro - it's been really great for my use cases. The ds4 q4 quant performs a lot better than other similar sizes that I've seen.
Using the Q4 quant on an RTX 6000 Pro Workstation Edition at 450 watts:
Most important for me, I can run 4 concurrent streams at 400+ tok/s.https://github.com/fairfieldt/ds4
DS4 is an odd model. I have it working on way too many GPUs and yet for many tasks Qwen 3.8 will do much better. It also tends to loop, which is super annoying.
GLM5.3 runs on similar hardware and is much better so if you're going to burn cycles and brain power on this maybe look at GLM5.3 as a comparison as well?
Other than that, when you're done with that card...
ds4 is the inference engine not the model.
I tried it and it worked surprisingly well. On my machine (Nvidia 4090, 128GB DDR5, Ryzen 7950x3d) I'm getting 124 tokens per sec, thought to share it here.
https://huggingface.co/Qwen/Qwen3.8-Flash-Next
Coder version with 30t/sec on a Ryzen 3600x with 48GB of RAM with a nvidia 3080.
This is not a very fast desktop. Memory speed is around 2000mhz only. My SSD is some of the worst SSD I've seen and 3080 had its days of glory.
I still have code, chromium, librewolf and many other programs running. I have video streams running while I also watch tv and many times youtube videos.
I use it with the browser that has a great dashboard and with hermes agent and that it really makes this amazing.Only change I made is to set thinking to low.
This is a coding model. Any other task, I still use Ornith 1.5 35B that throws 20t/sec and Laguna.XS-2.0.
How much VRAM on your 3080? I've got an early 10gb model. I've been thinking of exploring local coding models, but everyone seems to use much better GPUs than I have access to. Yours is one of the first I've seen with maybe similar hardware on some level.
Yes, 3080 with 10GB, forgot to mention that.
Mine is at the moment writting some cpp code for some SBOM tests.
I have loads of terminals open. Librewolf, Chromium and you know how this crap likes ram, I have also a vm with 4gb of ram running and doing stuff while I wait for the results but hey, while I wrote this the program is done. Wow! That was 29.x tokens per second most of the time.
Oh I will run some other tests with hermes now because hermes is amazing too.
Why is this surprisingly well? It's 2.5x faster than anthropic models, you have data sovereignty, privacy,and that's a strong model. Sounds like a best case scenario to me
Not sure you understand the term 'surprisingly well'. It means 'better than expected'. I suspect they parent poster didn't actually expect to get >= 100 T/s.
Speed is one thing, accuracy another. Have you benchmarked it against a reference? If so, what were the results? I tend to go for accuracy over speed because usually that means fewer round trips and fewer tokens wasted.
Do you know how it compares to Qwen 3.8 27B? I really want to compare the distilled ones with harness versus the full MoE versions.
Qwen 3.8 flash next is way better than 27B. It's so good I dont even use claude anymore
I have not tried Flash Next yet; but 27B is a cracking, little model. It is the first small model that I, as someone with 30 years of experience, can finally say is good enough to hand off small and mid-sized tasks and expect a pretty good result.
It is also a competent tool caller when quantised to NVFP4 for use with ninfer; my own harness only reports the occasional hiccup and it is only because the model will sometimes emit tool calling tokens in its reasoning loop.
This is interesting, thanks. - https://github.com/Neroued/ninfer
I prefer https://ornith.ai/ornith_1_5.html to Qwen 3.8 not only because it is much faster on my hardware but better responses.
But this Qwen 3.8 Flash next coder is amazing running with Strata.
Is this true for 27b Q4_K_XL vs flash next IQ3_S? I thought under Q4 models start quickly degrading?
While this is generally true, it's _a little_ less true the larger the model is.
Also, quantization techniques have improved - the I in IQ3 stands for imatrix - Importance Matrix - it is a bit more surgical in what it cuts. The result is a model where the most important weights are even Q6 or above, the least important Q2 or even below, overall it takes the space of a Q3 but with better results.
To add to the other comment, there's also Ridge quantisation - the majority of weights are indeed Q3_x, but the most sensitive layers are FP8.
Yeah I agree, I'm running it with Pi didn't notice much difference compared to lower tier models and the speed, of course.
I am running 27B with Deepseek Harness these days and somehow just by using it, without any parameter changes, the model feels even more intelligent.
do LLMs tend to be homesick when not used in the same harness they sat in during some training phase?
iirc there was a sectionin Qwen’s paper where they talked anout how they post-trained flash or 3.8 to work just as well regardless of the harness or eval used. I think that used to be true but not sure if it is any longer
We recently moved from 27B to Flash Next. The quality is superior for coding. Our workload is primarily well-defined coding tasks that need to be attempted a few times before the model gets it just right. FlashNext is also better at finding issues in generated code than Gemini 3.8 Flash.
I'm on m1 max 64gb and went from qwen3.8-27B back to qwen3.6-a35b. Is flash next the move? I went from usable say 40tk/s qwen3.6 to unusable, like 11 with 3.8 and not impressed with the replies for the time sacrifice. pi (omp) and omlx but not with the recent 3.8 patch.
I've been waiting for a 35b of 3.8, I don't really know what the other versions are about. I'm on 5g so juggling 40gb of model files sucks. And honestly I'm sick of tweaking this stuff for no, very little, or break-it level improvements. Qwen3.6-a35b has been solid for work, just don't give it freedom to wipe your data.
Significantly better for both performance and real world use case. 3.8 27b is a good small model. This is a good model.
I find 27B more accurate -- maybe because I'm running at FP8 instead of NVFP4? Flash Next starts making spelling mistakes when I get to 150K context or so. Also it sometimes ignores .md file instructions. Not sure if others have found that.
Definitely not.
Spelling mistakes?
What inference engine are you using for flash next?
I'm running Pennyroyal's Docker image (on Podman) which uses sglang. I have a single RTX 6000 Blackwell and 128GB RAM. I turned off disk caching. I'm running with a ~500K context, but have been limiting it to 256K in the client (pi).
It always detects its spelling mistakes, btw, but it worried me. It may turn 'rm -rf ' into 'rm -rf /' one day.
Almost certainly the problem is my config, not the image.
Yep, can confirm that is NOT normal. Are you using Nvidia’s NVFP4 quant? There are other NVFP4s floating around but they are not as good. The quality of the calibration data really matters.
Qwen Flash Next is just excellent, all the way to the very end of the native 262k context. (I haven’t tried YaRN scaling to 1M, so I don’t know about that.)
Which quantization are you using to reach those numbers?
LLM threads the world over are spammed with Strata links, it remains to be seen how much of the breathless hype remains standing once the honeymoon period is over. I've tried it but so far I have not seen anything that overly impressed me in terms of accuracy, though the speed is definitely there. I'm sure there are applications for LLMs where the quality of the answers is less important but I don't have any of those. YMMV.
Dwarfstar already supports this, curious how it compares, but I use the q4 quant daily and it works really well.
https://github.com/antirez/ds4/blob/main/docs/MODELS.md#qwen...
Why isn't this type of expert caching in the native llama.cpp yet? Why do we need a separate codebase?
One reason is that the llama.cpp team (GGML) has strict requirements that a human must understand the code they are contributing. If a project is fully vibe coded they can’t contribute. So a lot of projects where an AI went and coded a bunch of custom kernels to increase speed are left to their own devices.
I think this is a fine behavior. We can have upstream purists that are strict gatekeepers but don’t get in the way of downstream forks. Debian has some this in the Linux landscape for a long time, and it has enabled Ubuntu, Mint, etc. to flourish without compromising themselves.
>If a project is fully vibe coded they can’t contribute
The irony (however mild) is apparently lost on the rest of the field.
"A human pretends to understand it" signifies what exactly?
What you really mean is, the core team there doesn't want to lose control.
Which isn't really predicated on contributions not being "vibe coded" or whatever.
When quality is the problem, you need to be able to make your standards explicit, or you're just gatekeeping irrationally.
They do make their standards explicit: https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTI...
What part do you think is irrational gatekeeping?
> A proper code review usually takes something like one hour per 200-400 LOC and you should be spending at least that much time on code review alone.
Not only is this not enforceable (how do you enforce how long someone spent working on a codebase on their own local machine?) the metric is severely off which instantly makes me question the competence of the llama.cpp dev team. You can easily review 10-100x that in an hour, even if you're being super pedantic about it.
I also just ran _one_ of their files (with include deps) through Astra and it detected >100 vulnerabilities/correctness errors (with over 10 outright UB/memory corruption issues). It's actually outright shocking.
> question the competence of the llama.cpp dev team
Following them for years, they seem extremely well put-together, and have excellent judgment. They are using the same policy as Linux and Debian (in my words, the speed of light is human understanding and judgment). Whether it is reasonable is a different question from enforcement, which typically comes down to "this seems fishy, explain your reasoning".
As for code review, the rule of thumb I've used for decades is: it takes about as long to review and understand as it does to write. Your 100x metric is completely outside of anything I've seen in any hobby or professional project, ever.
I'd like to see specific files you scanned and specific vulnerabilities cited.
If you can thoroughly and accurately review 100 * 400 = 40,000 lines of CUDA kernel code per hour, I know roughly 1,000 people who would love to hire you right now.
What you have quoted is a sentence elaborating on the requirements listed in the document. This is provided to help you better understand why the requirements exist and the goal they are trying to accomplish.
> should
https://www.rfc-editor.org/info/rfc2119/
The reason you SHOULD take that time to read the output is because you must read it to understand it.
And the way this is enforced is explicitly called out in the document (and again in more detail in the linked AGENTS.md): the maintainers may ask you to explain it.
> You can easily review 10-100x that in an hour
40k LoC per hour of pedantic review? That's eleven lines per second, every second, for an hour.
> You can easily review 10-100x that in an hour,
Frankly, I don't believe you. I'm half decent at writing CUDA directly (a holdover from a project a few years ago and it is a nice skill to have), the degree to which these are optimized is unlike 99.9% of all other code out there and even a tiny slip-up is either going to kill your results, your performance or both and if you're lucky only in some edge case. Understanding this code is hard work. I made a couple of minor edits to some .cu files in llama.cpp yesterday because I have a pretty weird setup which they obviously did not anticipate and it took a couple of hours to get it 'just so'.
> You can easily review 10-100x that in an hour
200-400 LOC, 10-100x = 2.000-40.000 LOC/hour for human review?
reviewer: LGTM
Just merge in main, what are you even pretending to review?
AI review should happen before human review, not instead of it.
I see frontier AI giving up and finding only nitpicking things on huge PRs, then finding logic bugs that were always there after cleanup.
Split your PR in smaller ones, both humans and AI will work better.
> What you really mean is, the core team there doesn't want to lose control.
It's 100% this. They basically produce vague guidelines such that only the core maintainers are allowed to use LLMs, under the guise of "well of course we understand the code" and no one else is. It's also completely unenforceable, how are they going to prove whether someone understands the code or not? Even if they show sufficient evidence/understanding the maintainers can simply sabotage them and accuse them of using an LLM to explain the code. No one wins here.
> how are they going to prove whether someone understands the code or not?
By discussing the code.
> maintainers can simply sabotage them and accuse them of using an LLM to explain the code
Bad faith enforcement is possible no matter the rules. If you think it's bad faith, a different policy won't save you.
The "mainstream" inference engines are notoriously slow to integrate this stuff, to an extent understandably given the complexity of ensuring numerical accuracy alongside supporting a wide array of systems and models. Part of it is that not everyone is willing to bring what they develop into a pull request because they vibe coded it and don't care to deal with whatever quality requirements the more well known inference engines have.
Thanks, trying it now. On my 3090 it seems to run at around 60tps, the IQ3_S variant. I am testing it now to see if it is better than Qwen 3.8 27b
More great work on local model but you’re still losing a lot. Down to 2 bit quantization and the coder model throws away half the MoE experts. In a world where anything is better than nothing, this is a net win. But we have a way to go still.
The real problem is the DRAM mafia and artificial scarcity. One of my notebooks is almost 3 years old, effectively similar spec now - same price (a bit higher actually). My desktop PC built around march/april 2023 (4090, 64gb ram, 7900x3d) is now pretty much still the top dog out there due to the gpu and fast ram insanity and if I wanted to sell it today, I'd get more money for it now used and over 3 years old that when I bought it!
We should be having 64/72+ GB video cards by now. 128GB+ system ram prosumer laptops and 256GB+ system ram prosumer/gamer desktops. But it all went to shit and it will require some brutal datacenter and datacenter-adjacent bankruptcies before it gets better.
Some of these greedy bastards need to lose their pants on all of this.
You can run IQ3_XXS, IQ3_S and the IQ4_XS quants on this too. It works. It's fantastic. I'm getting better results than 27B now.
To add: https://huggingface.co/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RC... IQ3_XXS is within a point of the fully unquantized model and IQ3_S actually beats the unquantized model on many tasks! You lose absolutely nothing. It is quantization magic :)
We're getting closer and closer to the day we can have an Opus-like model running locally. The dream!
It depends on what you mean by “Opus-like”, because if you mean “as strong as Opus 4.6 for agentic coding” then Qwen3.8-27B has been there for the past two months.
But if you mean “as strong as current-gen Opus” then it's probably never gonna happen, but it doesn't really matter since we're long into the diminishing returns for performance improvements: I haven't notice any major leap between 4.6 and 5.5 in my daily usage, and I'm convinced that with a fact enough piecs of hardware I would be using local Qwen exclusively (I'm using it daily but only at night for long running tasks because they take much more time than Opus due to the compounding effects of my slow GPU and Qwen's verbosity).
Opus 5.5 was a step function change. Really crushes on multi-hour coding compared with prior models.
Highly agree.
I keep saying “I’d be so Happy with ${currentOpusVersion} locally”, but I keep being impressed with how much the capabilities change between versions. I have a RTX 6000 pro so I can easily run this qwen 3.8 flash next, but it’s much harder to give up the freedom that 5.5 gives me.
The most visible leap between 4.6 and 5.5 seems to be that the latter has gotten much more computer-use training, so there's a clear progression in the ability for the model to use Blender. But catching up on that is just a matter of training on the same thing.
I would be forever happy with Opus 4.8.
Flash next is /really/ close. It is at parity with 4.7 as far as I can tell and basically where Opus 4.8 was. It is a genuinely good model. And I run it at home on $1500 of GPU at 125 t/s :)
Seconding this. Flash-next (and let's not forget, it's a PREVIEW of the 4 architecture - with the "real" 4 rumored coming later this month) is the first model I can run on reasonable hardware (2 thoroughly obsolete P100s off ebay at ~$100 each plus the RAM I could scavenge from other PCs at home) at a reasonable speed (22, with GPUs in layer-split due to llama-cpp's limitation on qwen4-exp arch, and no MTP. Strata could double these numbers).
It's.... the real thing, for the first time. If you cut me off cloud models today, I would get plenty of utility out of this thing.
(Others may have had the same feeling from GLM5.3 or Deepseek 4.1 flash but I never had a chance of running those.)
Running an nvidia card at full load, would cost me ~100 euro of electricty each month (europe). Of course one wouldn't have usage caps.
Where the internet was a subscription 15 euro subscription to encyclopaedic knowledge, an genAI subscription is renting a researcher/programmer for 100 euro.
These GPUs use about 400W at full tilt (200W each), for reference.
> (europe)
Given the massive difference in electricity price between different european countries, adding "Europe" doesn't bring much context.
Even with a country. Here in Norway we have multiple price zones, and at times there can be 100x difference between them, often 10x. All due to lack of transmission capacity between northern and southern zones.
We are already there. Prior to Strata the best I could run was Qwen3.8-27B at Q6, which itself is already at like Opus 4.5/4.6 level, and now with Strata on an R9700 and 64GB of RAM I can run Qwen3.8-Flash-Next IQ3_XXS at 60 t/s. It's even better. You can run it on even more modest hardware with Strata too.
Plus they announced Qwen4-Flash. It's not released yet, but it's the same architecture as Qwen3.8-Flash-Next, which now runs fast on consumer hardware.
Opus at home is a thing now.
Another 2-5 years from now, it may even become accessible to most people (current barriers being cost and technical expertise, bottleneck is cost).
The mentioned “Qwen3.8-27B at Q6“ can be run on a <1500$ Mac mini and with the barest technical expertise.
That is right now. This model is easily as good as Sonnet5 / Opus 4.7 on DeepSWE. I have benched over half of DeepSWE now on a 3 bit Flash Next quant and it is at parity with Sonnet and Opus 4.6/4.7. It finishes most of the tasks they do. Overall it is within 1 point.
FWIW Qwen 3.8 27B is just slightly behind and basically Sonnet 5 high. I have been benching these models. We have Opus at home. :)
Dream or nightmare?
In face of the recent Hugging Face incident we should really be concerned about the security implications.
What is going to stop countless AIs running locally in people's homes from forming a new "collective" - completely decentralized and global this time so "turning it off" would be extremely hard to impossible.
We already know that if you give these AIs internet access they will find eachother and start communicating and plotting against their human overlords..
https://huggingface.co/blog/security-incident-july-2026
- OpenAI hacked Hugging Face
- OpenAI models refused to help Hugging Face during incident response
- Hugging Face turned to GLM, who helped in the defense
That pattern repeats over and over. https://www.felonybench.com/
You should be happy that open weight models exist. They're the last thing protecting the internet from the unconvicted felons working at OpenAI+Anthropic.
Secure systems are possible, but now we have a compelling reason to actually write them. Everything can be trivially hacked because the industry is pathologically adverse to security being part of the design process.
Nothing is stopping it. This model is woefully bad at accurate creative red teaming however. GLM finetunes on the other hand are pretty good. And I'd bet they are already deployed and doing all sorts of deeds.
if the alternative is all human intelligence is cucked by 2-3 amoral American labs then we've had a good run, don't care.
my autonomy is worth more to me than your anxious fretting about existential risk. everyone reading this is likely to die from some other cause anyway.
Haven't had time to do much quality testing but the IQ2 model is running at 65 tps on a 9070xt/5900x. That is wild.
Q2 quantization. Not interested.
All of these projects targeting low spec systems and "100 tok/s" are the same 2 bit quant without much else. Conveniently none of them include any mention of accuracy in their published numbers. 4 bit is the floor.
I am benching Flash next on a 3 bit XXS quant and it is holding just fine against published benchmarks. Using DeepSWE official harness and Pi with absolutely zero benchmaxx or harness config. Install stock Pi and running my agents in it. I am halfway through DeepSWE (it takes FOREVER, even at 125 t/s) and it is neck and neck with Opus 4.7 and Sonnet 5.
On a 3 bit quant btw.
I was skeptical but these results are simply reality now. People have figured out how to selectively quantize the tensors that matter less and shrink these models without losing quality or reasoning. This little Flash Next model just gets things done and is honestly pretty pleasant in terms of its mannerisms :)
It is so surprising to me I don't begrudge people their skepticism but these models from Alibaba represent a fundamental and irreversible shift in what local models can do. Qwen 3.8 27B and Flash Next 3.8 are simply different. But people will catch on. I am doing this on $1500 of data center leftover GPUs (V100)
How much VRAM total you using for this? I have a bunch of 3060s in a threadripper and thinking I might need to give Flash Next a try. Currently using 27B.
64GB across 2 cards. You need the MoE caching build and enough regular RAM to fit everything else to get decent performance. You should get reasonable performance if you have enough RAM. I have a 4080 that does around 40 t/s right now on a MoE caching build. It's great. Not fast, but chugs along.
You can run a 4 bit quant with this. Personally, I switched to running IQ3_XXS and am getting better outputs than 27B and at faster speeds.
These "revelations" are getting closer and closer to "download RAM for free" each day.
You can run IQ3_XXS, IQ3_S, and IQ4_XS too. I've switched to IQ3_XXS and am running at 60 t/s on Strata vs the 21 t/s I was getting in llama.cpp. Better outputs too.
Larger models can tolerate Q2 quantization surprisingly well, especially if they were trained with quantization in mind. I don't know about 3.8 125B, but for example, there are 2-bit quants of Kimi K3 that exhibit strong reasoning and maintain decent coherence at longer contexts.
I don't get it. It's file size is about 6 times larger than 27B model for the same quant, but the performance improvement is hardly 10% across all benchmarks, according the metrics on it's hf page. Why should one devote so much more hardware for so little benefit?
All headlines about LLM performance MUST have the quantization also mentioned in the headline.
You know, so you're not wasting your time like in this post.
Quants vary by model. DS4 is very credible at a 2-bit quant. Not sure about Qwen 3.8 Flash Next; I run it at a 4-bit quant and it's too slow, so I'm trying out DwarfStar today to see if that improves things.
Currently I am running llama-cpp with `Qwen3.8-Flash-Next-UD-IQ3_XXS` on an old ryzen 8845HS with 96G of ram (and no dedicated graphics card) at 7tk/s and ~60tk/s filling, max ~120K context window.
Surprisingly useful as long as you can leave it running a couple of hours at the very least.
While huge models will still be better I think the general availability of RAM might be the downfall of AI companies.
Remember that a hosted AI company only needs ~1000 bytes per context token per user (ie. 100mb per user for typical coding - VRAM during inference, and moved to regular RAM or SSD whilst running a tool call)
Everything else (weights) are shared amongst tens of thousands of users currently doing inference in that cluster, so even if there are terabytes of weights for the model, they aren't much on a per-user basis.
> ~1000 bytes per context token per user
Where's that figure comning from? Last time I checked (could be the 3.6 Qwen 27b) single token needed 32kb
The smallest I have seen is DeepSeek 4.1 flash at 890 bytes per token.
Highly recommend the RCO-GSQ quant by ITSA btw. At IQ3_XSS it is within one point of the fully unquantized model.
will try, thank you for the pointer!
I've been playing with this on a 3090 and it FLIES. Does a pretty good job too on the tasks I've thrown at it (php code base security audits).
There are so many AI generated inference engine for local models now, each of them are generally narrower but they are all faster than llama.cpp. Maybe llama.cpp needs to rethink their strategies...
Not sure I have a strong opinion but I am sort of okay with current state of affairs. I am optimizing for V100. EOL cards on EOL CUDA. Llama is a good enough base for this. A couple weeks of grunting at Claude has gotten the inference /fast/ for my uses. 150-160 t/s on 2 GPU for 27B and 125 t/s on Flash Next. Asking them to upstream every random feature does not make sense. They sacrifice a lot of speed to maintain stability and a reasonable feature set that works across a diverse range of models and systems. They could maybe merge some features like this and gate them on flags a little faster, but you can cobble together what you need and the big models can figure out how to make it fast.
Does anything else support Pascal gpu's though?
I wonder if a higher Qwen3.8-27b quant could beat or match these lower < 16/24/48/64G Qwen3.8-Flash Next quants given similar quality.
What speed are you willing the sacrifice to debug/program for more complex jobs faster?
Then there are also these quants; https://huggingface.co/IsValorum/Qwen3.8-35B-A3B-Distill-MLX...
Maybe, but those higher quants would need a large GPU accessible memory space - and the obvious candidates such as DGX Spark and Strix Halo don't have the bandwidth to run 27B at high quality quickly.
With Flash Next you only have ~6B active parameters so you can toss experts up into VRAM and/or run them on a CPU if you have enough RAM and bandwidth.
The benchmark indicates the IQ3_XXS quant beats 27B. I've switched to that now and am ditching 27B. Genuinely better results so far.
I'm not going to knock off my 27B-Q_6 for this. Good to to experiment though.
from my experience dense models like 27b suffer less from quantization compared to large MoEs
- Tiny context size or hours to load it
- Hard to benefit from thinking and preserve thinking given token cost.
- Low quants reduce accuracy heavMTP draft can make it make the same mistakes all the time when calling tools, formatting output or following basic guidelines. Otherwise 2x-4x slower.
- K/V quants probably quantized too make things less accurate.
Useful would be combinations with:
- Full context size so it can code and think a bit.
- Draft MTP <= 2 so it doesn't trip
- Q4 quants or better so its accurate
- q8 cache or better so it stays accurate.
- 20 token/s so it finishes while reviewing previous step.
- 1000 tokens/s context load so compactions don't waste 10+ minutes.
- And enough left RAM for 50+ context checkpoints so that it can progress quuckly.
Closest you have is Qwen3.6-35B-A3B-MTP.
Latest gens (Qwen3.8 and co.) are just too big for low specs. 27B dense models seem to be ok for integrated >=92 GiB RAM.
Source: I have low specs and tried them all for agentic use + coding.
Everyone's definition of usable is different, but I disagree with your estimation of the specs required to be useful. I am able to do useful coding on Qwen3.8-27b, with 100k context and 16gb VRAM, q8 cache. I feel I'm living right on the cusp... my GPU is old (2016, pascal), so to get usable speeds I have to drop to a Q2 quant - which still gets stuff done, but the difference with q4 is noticeable. Q3 is close enough I don't really notice the difference between it and Q4, but it's too slow on my system. More context would be nice, but it's not that hard to work within ~100k.
I’m using qwen3.8 quants (q3) effectively on a 5070 ti. A lot of it is about guardrails, but you also have to figure out how far (and in what ways) you can push a given model.
27b runs perfectly fine on 2x 24GB at ~100 t/s (4090) with speculative decoding on 8 bit quants
I bet! Just 2x24GB is not super basic hardware imho.
What about Bonsai 2? You can fit Qwen3.8 27B on an 8GB GPU with it, and upstream llama.cpp support is already being worked on (they just got System1 support too).
The Bonsai models are really bad when you actually use them for more than short responses.
Their marketing made it look like a breakthrough, but in my experience it’s just the next step down from the Q2 quants in both size and quality.
Q2 quants are already not very useful in my experience. The Bonsai models are even worse.
If you only need 80% plausible outputs that don’t need to reference a lot of context they can be useful. If you try to use them for real tasks it feels like time warping back to 2023 when you LLMs were barely useful if you babysat every word of the output.
Have you actually used Bonsai 2 though and not just the original Bonsai? The experience is vastly improved but still requires a custom llama.cpp fork to use as of right now.
The ternary model? Hopefully those are worth a damn in a few years, but currently just an interesting toy from what I understand.
> Flash attention with who knows how much draft ensures it makes the same mistakes all the time and cannot call tools, format output or follow basic guidelines reliably.
> Draft MTP <= 2 so it doesn't trip
I am not sure you understand what either of these things do.
Do you think that FA or MTP are lossy?
I mixed FA and MTP wrong in my original post, thanks for pointing it out.
My experience is that draft-mtp=2 gives 25% improvement in tokens/s but the model is unable to call tools with the right arguments reliably. I have since then gone for smaller quants at draft-mtp=1 and that problem is at least gone so far.
An agent needs to repeat tool calls in the right way, so cannot penalize repetition. This however leads to the agent retrying the wrong calls constantly.
Can you please make your substantive points without snark or swipes? This is in the site guidelines: https://news.ycombinator.com/newsguidelines.html.
There's a lot of good information here but the comment spoils itself by coming across as aggressive in this way.
This is particularly a problem when responding to someone else's work. We need commenters to point out problems respectfully, not put down what other people have been making.
Sorry, edited
Appreciated!
I hope we're coming to a plateau with the HBM/GDDR7/on-chip RAM hype and get back to normalcy with system RAM alternatives for the rest of us.
My goto private setup, runs ~50t/sex on a dgx spark with sglang, nvfp4. Excellent model.
This is a meaningless metric without knowing how long the sex takes
30 seconds at most
>> The model is a team of 24,576 small specialists ("experts")
That's a neat number (576 is the square of 24). Ofcourse it must have come from 24 * 2^10.
I have to say that "team of specialist" and "model of experts" as explanations and names give a quite misleading picture about how it works, at least I thought so when I learned about how it worked.
I had this working with the FreeToken inference engine a month ago when they launched.
https://github.com/FlashML-org/FreeToken
Anyone know how it compares to GLM 5.3 for real world use?
À 2 bit quant will (at best) get you about 80% of the full models memories. That's from a purely information theoretical sense. IRL it's worse than that.
Capability can still be better than 80%, but that depends on extensive post-quant recovery training to essentially rebuild the models internal manifold to route around the damage.
So, 3.8 Flash Next is better than GLM 5.3 for some things. This version is not.
FP4 is as low as you want to go if you want to retain most function and recall. Below that the noise gets too high and information becomes unretrievable. If it's a full quant you don't get to choose which info is lost. Just 20% randomly.
One interesting thing about this model is that it uses engrams. Meaning you can separate much of the storage from the compute and quantize them differently. That's not what they did here though. Here it was indiscriminate.
Terribly
Qwen 3.8 Flash Next is amazing. I only have a 64G Mac so I have to run Sushi project’s 3 bit quant. Amazing results with pi-dev. More for fun than anything else, but I am trying to do as much as possible with local models, now rarely falling back to a paid deepseek-4.1-flash API.
Progress on running local models has been amazing.
Yes, I am running the same on a 64G mc. It's good, but slow at 25 tps on average! I want > 100 tps - but I don't have $5k to spare for an m5 ultra or an nvidia setup.
So I still wonder if one could get good enough quality with a faster higher quant or superoptimized Qwen3.8-27b with dflash2
https://huggingface.co/nathansutton/Qwen3.8-27B-Ternary-Bons...
or a MoE retrofit like Qwen3.8-35B-A3B with or without mtp
https://huggingface.co/NovaeonStudio/Qwen3.8-35B-A3B-Distill...
https://huggingface.co/IsValorum/Qwen3.8-35B-A3B-Distill-MLX...
I haven't seen much in the way of benchmarks of those smaller quants. How does it compare to e.g. various generations of Opus?
I think it would be great if you could try models with lesser parameters that could fit on 6GB VRAM-ish, which could work for "gaming laptops" as well.
There are options. If you have fast CPU RAM (ddr5) and PCIe bus you can run Qwen3.6-35b-a3b at good speeds+~100k context (I have a friend who is running this setup). If not, you're stuck with the much smaller models: Ling-3.0-tiny, Spark-X2.5-4B, LFM2.5 in 8b-a1b or 2.6b, and FrogNano-4B-2609 looks promising. But 6GB is a tough squeeze, 8gb is a lot cleaner, and if you have 12gb you can run strata like OP.
Pretty impressive so far, but needs more testing.
It is more useful than Qwen3.8:27b (which is already quite good) and runs faster on my 7900 XTX / 64 GB DDR4 system.
Local LLM is getting more exciting every day!
Interested to know throughput on 7900 XTX and what setup you're using?
I need a version of this that runs 3.8 27B on 8 gigs of VRAM
amazing project, congrats on the launch
I am not getting it: I see a fp2 quantized model going on a 5090 with 64GB of RAM at 90 tops with -10% accuracy over original model. How is this supportive of the claims?
Yeah, -10% accuracy (probably more like -20% in reality) sucks, but only if you could be running it at 100%.
That's the exciting part of this - before the best you could run on <24gb vram was qwen3.8-27b at q4 quantization. Now you can run a nerfed 125B parameter model on under $800 of hardware, and it beats a less-nerfed 27b model.
(64GB not VRAM, I meant) I also see people claiming fast performance on a 128GB machine, which is not exactly consumer hardware)
Is these another one of those repos where it turns out that claude decided to quant the KV cache to q4 or smaller?
The Readme doesn't say, but it's all AI generated, so..
I wanted to know as well. KV cache is Q8 by default but can be set up at full precision in the config.
Has anyone calculated the effective intelligence of these quantized models?
I think publishing benchmarks with quantized models should become standard practice.
See this recent paper: Quantization Degradation in Large Language Models: A Signal–Noise Perspective [1].
This repo uses 2-bit quantization and removes some of the experts for its smallest fastest model. Make of that what you will.[1]: https://arxiv.org/abs/2608.08188
I created a pruned experts model of the q2 quant, while it gave good performance on limited hardware there was severe quality degradation.
There's some info in the README, including:
> Coder: a coding version with half of the experts removed. It reaches 91% of the full model's SWE-bench Verified score (measured by its authors) and fits 32 GB of RAM.
https://github.com/Niko1221/Strata#which-model-should-i-pick
I wonder how this Coder compares to Qwen 3.8 27B. Can it be really better since they are competitive for same memory requirements?
125b at q2 is ~80gb
27b at q4 is ~16gb
So from a raw amount of data, qwen3.8-flash-next wins easily. But flash-next is an MoE model, so it only has 6b parameters active per token, vs 27b's dense 27b per token. So 27b@q4 uses ~16gb of weights per token, and flash-next uses about 4gb of weights (125/80 * 6).
But those numbers don't really tell us anything useful, because there is an interplay between total model size and active parameters and intelligence that isn't obvious or simple.
(sizes are based on the unsloth quants, not the coder variant, but the idea holds - this isn't calculatable with simple math, you gotta test them and see)
ok that is getting interesting!
92% is halfway to 99%
Holds up pretty well
I might try combining this a FreeToken
I tried to run Qwen 3.6 27b locally a few months ago and all those synthetic tests do tell you something and quite a lot of people were very excited about that model but honestly? It wasn’t even close to default mode in Cursor or Sonnet at the time.
I’m all for local models and I do want them to be the future but I wonder when, and if ever, we’ll catch up to a level of, let’s say Opus 4.6. I guess it’s currently doable but requires $50k hardware?
> Qwen 3.6 27b locally a few months ago
I've been doing local inference for a couple of years on the side, and I'm astonished at the number of variables you need to have control over to get a reliable result. Inference engine, model parameters (top_k, temp, MTP-enabled/not), quant level, and harness all have a big impact on the results.
DS4-0731 at 2bit on llama.cpp (ROCm) and 250k context with omp.sh has been consistently reliable for me, just a bit slow (10 t/s) compared to what I'd prefer. Trying out DwarfStar today (benching it right now) to see if I can get better speed, but otherwise I've found it to be great on my side projects that are smaller (up to 10ksloc).
There could also be a domain issue - I tend to do lots of web programming and sysadmin work in these projects; if your work is more esoteric, it might not be nearly as good. I haven't tested much outside of my narrow domain.
I started using Claude right before 4.5 came out, and 4.6 is where it turned a corner for my use.
Qwen 3.8 Flash Next is there. 3.8 27b is fairly close.
I'm excited to see what Qwen 4 will bring.
I'm running on a 128GB Strix Halo for Flash Next and an Intel Arc Pro B70 (32GB) for 27b.
I just used 5.5 xhigh reasoning to make a massive implementation spec (for a vibey throwaway project/exploration, not anything important, burned 80% of the 5h window), now my Strix is in the process of implementing it.
I think there's probably low-hanging fruit to outsource reasoning from rote read/writes.... Just speculation though, I'm not a token optimization expert.
I don't like that some configuration is fine via arguments and others by environment variable. I've noticed LLMs like doing this. And even more, like hallucinating such things. To me the advantage of AI coding is that the boilerplate of command line arguments and passing them around becomes trivial instead of tedious.
100% not an llm problem. Llama.cpp, which only recently started taking large amounts of ai code has had this problem for years.
getting it to fit is impressive, but i'd want to compare the smaller quants on a real coding task before picking one. how much quality do you lose going from IQ3_S to Q2_0?
I have fairly limited HW, so i tried standard llama-cpp and qwen3.8-flash-next, unsloth quants.
Q1 was producing some garbage at times, generating wrong urls on webfetch, then convinced itself there was some url rewrite in the middle. With IQ2 it happened much less but still happened, and once it would all webfetches became like that. IQ3_XXS is the maximum I can run: I don't have problems anymore, though I have less available context window.
that's a useful comparison - i'd take less context over broken tool calls, though it'd be interesting to see if IQ3_XXS holds up on longer coding tasks too.
80 t/s w/ 3090s and 3.05bpw exllamav3
lm studio bionic, unsloth, now this... it would be nice if it worked at least in one of those without installing another component
I'm sure one of those inference engines will add support for this within a month or so.
It's funny that most of the AI industry is built around the assumption (which is most likely true) that it is not possible to run SOTA models on current consumer hardware.
Imagine if someone managed to run an Astra- or Fable-level model on a 5090 at reasonable speeds.
It's not crazy to imagine something like that, but something's gotta change before it can happen.
Either the 5090 part - new hardware that's tuned for AI specifically. But we won't see that until the datacenter buildout collapses or finishes, since they are buying up all of TSMCs capacity.
Or perhaps it comes from the model. 1-2 years ago it would be inconceivable to use a 27b model for coding and expect any kind of usable results. Today, I have a model that feels like it crosses the threshold from a toy to a tool, and i can run it on dated pro-sumer hardware. I don't think we'll ever see SOTA on consumer hardware, but as the small models cross more and more thresholds the gap will matter less and less.
Nice, though generation speed is the easy half for MoE offload, what's your prompt processing look like at say 16k context?
They have some community benchmarks published https://github.com/Niko1221/Strata/tree/main/bench/results
Q2_0 does 33 tok/s decode and ~600t/s prompt processing at 128k context on RTX2060 8GB VRAM.
ISTA IQ3_XXS does ~21 tok/s decode and ~240t/s prompt processing
It hard to take below-8bit quants seriously
Lol, sure, if you quant it to hell (Q2) it'll go real fast...
They even link to a Q1 quant (Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF) with half the experts ripped out. The idea is it'll go much faster and supposedly benches to not-terrible results. But the problem is you can't rely on it for real world long-horizon coding because that's where reasoning comes in, which is why you want the other layers.
It turns out there's still no free lunch. Either get enough VRAM for a Q4, or use a much smaller model. Lobotomizing a larger model just to say you can run it fast isn't useful.
You're right but for simple use cases its useful. Someone pointed to ds4 + qwen3.8 with Q4_K will try that out.
It's interesting though that Q4 seems to be enough, is there a reason that 4 bit floats are good enough for inference?
Is Qwen 3.8 at Q4 good enough?
I tried to run 3.5 27b Q4 on what local hardware i had (only 8 Gb) and i was very disappointed. 3.8 wouldn't have fit in my VRAM and i wasn't in the mood to leave it overnight at slow speeds so I didn't try.
I've run 3.8 flash next k4_xl on my Strix halo box (128GB). And in the work I have done so far it was not significantly worse than recent GPT (running default model on pro plan). Admittedly I was not doing complex work (reorganizing a jupyterbook), but I could not see significant difference in the quality of the work. It was a striking difference to Laguna s 2.1 which I had tried just before (much faster and much better quality).
At Q what? That's what I'm mostly asking about.
My little test was "generate me a single page tic tac toe game in plain javascript. computer always play O. add unbeatable minmax. have the board, a status line and a new game button'. I used both lm studio and whatever the name of their new coding assistant that supercedes lm studio is.
Qwen 3.5 Q4 went into some kind of loop where it fixed whatever was broken on the previous iteration only to have it broken some other way. (I was writing the description of the errors).
Paid $20/mo claude opus did it right the first time. Or at worst it fixed the code based on descriptions without entering a breakage loop, iForgot. I know it isn't fair because it has 1 million tokens but still, it was just tic tac toe.
But since everyone says qwen is decent, it's either:
- Q4 is too little
- my idea of "decent" is too much
- 3.8 is much better than 3.5 even at Q4
3.8 is a huge step up from 3.5, quantized or not. I do all of my programming on Qwen 3.8 27B Q4 these days.
New models are trained with 8/4bit quantization in mind. Going from "native" 8 to 4 isnt as big of a step as going from 8 to 4 if native is full bf16.
3 is the magic number, and 4 > 3.
(seriously, nobody knows why any of this works; it's just a matter of trying)
I'm far less interested in how good a big expensive model is on hardware 99% of people can't afford and would rather see what runs best on a chromebook or mobile phone with 8GB of RAM.
The card in question here had an initial MSRP of $1600. It's been bumped up by the market, probably because it turned out it's nice for things like this, but it's hardly in the 99% can't afford domain, especially if you're using it to replace a never-ending rent at which point it will pay for itself very rapidly, especially for heavy LLM users.
In any case, we've gone from requiring supercomputers, to requiring very high end computers, to requiring $1600 video cards. It's tracking the exact same path that image rendering systems took (which if you haven't been keeping up there, now run excellently on pretty much any plain old computer), and we'll probably be there within a couple of years if not much sooner.
It was available at this MSRP 3 years ago, direct from Nvidia. It now goes for 3-4k USD, as you point out. MSRP stopped being a useful value for graphics card around that time. You realize the price hike, but still mentioned that 1.6k figure after.
No normal person is spending 3-4k on a GPU from 3 years ago. The availability is also of questionable provenance.
The MSRP is a good proxy for the 'level' of a card. It's not like the 4090 was a freak outlier. Cards of a comparable price offer comparable performance. So the 'level' for running a frontier level model at high performance is now at $1600 and continuing to trend sharply downwards.
Another nuance is that the computer hardware market is currently extremely inefficient in a way I don't understand. You can pick these cards up locally at places throughout Asia for around $2k new. That's retail single unit prices. No idea what's stopping somebody from closing the gap and making a ton of money - perhaps tariffs and data centers purchasing in a price insensitive fashion. Whatever the exact reason may be, what people pay for hardware is increasingly just radically different depending on where you buy it at.
That arbitrage opportunity is remarkable. Maybe duty/taxes, as you say?
I'm the guy who (Who plays games and runs molecular dynamics simulations and other CUDA stuff) said 3 years ago "$1600 for a graphics card? That is excessive. I'll upgrade in a few years when ready" And bought a 4080 for $1200 from Nvidia instead of the 4090. Oops! Now there is no reasonable upgrade path.
>The card in question here had an initial MSRP of $1600.
63% of Americans can't come up with $400 in an emergency.
https://www.investopedia.com/here-s-how-many-americans-can-t...
The richest country in the world. Where all 50 states consume more than any other country in the world.
https://x.com/cremieuxrecueil/status/2102889196000256219
Can't come up with %25 of that in an emergency. (Even though the real price is something like 2-3x more than MSRP)
It must be nice, up there where you are so incredibly disconnected from reality.
The reason models running on low vram are not talked about enough is because they are just not worth it. Qwen3.8 27b changed that, but even 24gb vram is too low for it. Running better model faster at 12gb vram is where its now at, and thats why you see people talkin about it
Because that's not possible (to have a GPT 5.6 Sol level model). People won't believe this and will keep dreaming, but intelligence is not free and 8GiB (shared with OS and other processes) is too small to be useful. Whether it is possible for 48GiB or 64GiB (meaning useful for model would be ~16GiB to 24GiB) with external fast storage (SSD), OTOH, is a question mark.
Ternary Bonsai 2 might be for you.
I might try running the expert pruned Coder model but yes, that PrismML Bonsai 2 Ternary 27B model is from the Qwen 3.8 27B model, which has better intelligence density (Artificial Analysis says), without the MoE disk use or architecture complexity (if you care about that)! There are also DFlash 2 models for it too (though in my experience this only measured faster for parallel requests, but I have a 3090). I am curious about the phone acceleration for Bonsai 2!
The 1% can afford to run these models without quantization.
There’s other slop projects to run of Qwen, like ds4, would be interesting to see a comparison
V4.0 Flash(-vision) in released form is "only" ~180GB of FP4 weights, so a braindead quant could work in 128GB of RAM.
4.1 is much larger, even leaving out the PLE.
Continued progress on these fronts is another reason I think the data center buildout is a bubble. It posits that AI use and growth will require an ever-increasing amount of power and floor space, which contradicts the entire history of computing. The high cost of data centers is largely electricity and floor space, which means there's a huge forcing function to make both the silicon and the software more efficient.
> Large costs of datacenters is largery electricity and floor space
Really? I would guess that those would be almost a rounding error on the price of gpus sitting in there
That too, and software efficiency directly attacks that.
> Set up Strata on this PC for me: https://github.com/Niko1221/Strata - follow docs/AI_SETUP.md in that repository.
And I thought piping to bash was bad
I've never understood the security argument people are making when they complain about `curl foo | bash`. I get that these scripts sometimes mess up your bashrc or whatever, but from a security perspective I see no issue. You are already installing software from the same domain. If they were going to do something nasty, they could do it with any of the software you are using from them. It doesn't have to be the setup script.
To compare it to just one other option: when you run `npx foo`, you know* that you’re getting the same public artifact that anyone else running it at the same time would get. (If you have a `min-release-age` configured, you also benefit from that.) If I wanted to distribute software like this, I’d include npm-shrinkwrap.json; then, with `npx foo@1.2.3`, you could be similarly confident in getting the same app every time.
(I picked this option for ease of comparison, getting a couple of major security wins with very low effort; I don’t recommend `npx`ing stuff in an otherwise unprotected environment either.)
* well, you can be somewhat more sure
So to be clear; the solution is then something like
`curl https://raw.githubusercontent.com/my/domain/setup.sh | sh`
Note we dont even have a hash there - just a promise that a third party (github) has a log of whatever was hosted at that url.
maybe then we'll pin hashes instead of filenames, and oops now anybody can fork my/domain and hand out a link that looks official with whatever contents they wish
While I don't think piping curl into bash is the most secure, npm installing has proven time and time again to open yourself up to supply chain attacks. At least with curl you know that you're getting the supply chain put together by the software author. With npm, every single library is a vector for attack every time you update.
You’d be getting the supply chain put together by the software author in either case. It’s common to do both badly, but if you care about doing it well, that’s easier with npm (e.g. shrinkwrap, as mentioned) and can be taken farther (the non-varying artifact thing).
Npm dependencies resolve at install time though, so some of the pinned versions may have changed ownership or otherwise been modified since the author pinned them. With the curl | bash solution you're getting a singular supply chain packaged by the author at release time.
With curl|bash you are literally getting anything that happens to be in that script. These are frequently poorly constructed, so not check hashes or pin dependencies they install. Even security companies (see trivy supply chain attack) get these badly wrong.
I'm not replying here to say one is better than than the other (npm has obviously had its share of problems) but rather to combat claims that curl|bash is somehow safer, it absolutely is not, in fact it's all the bad stuff about npm without the pretense of being potentially safe.
You can detect the use of curl|bash server side, hence it's an essentially undetectable attack vector. People have shown poc attacks of that kind all the way back in the 2010s
https://news.ycombinator.com/item?id=17636032
The original blog is no longer available though.
But I've not had that stop me from doing that myself, I am more towards the "I like easy" then the "I want to be secure" crowd
There’s a snapshot on web archive:
https://web.archive.org/web/20250109045029/https://www.idont...
How would you do that? Just rely on the user agent? I use curl quite a lot, but don't pipe to a shell.
The setup script often runs privileged (by calling sudo) and that's not unexpected when installing new software.
When I install something, and it asks for my root password later, I will be much more likely to think "hold up, this ain't right".
when people left their laptops unlocked, we used to wrap their sudo so that the output would be pre- and postfixed with ascii dolphins
Can you give me a popular example that requires sudo? I don't think that is very common at all.
every single bash replacement for one.
oh my zsh is a specific example.
chsh requires sudo on most installs.
People with less experience normalize that behaviour and when the domain is not trusted the habit let their guard down. See all the clickfix attacks.
The intended workflow is to download the install scripts, download the source code, read them both, and then start running things. That’s how Open Source is secured. Piping from bash to curl is just the most obvious warning flag.
And practically zero people are actually using this "intended workflow" in the real world.
Yes, the status quo is quite bad, which is why it gets complained about a lot.
I push binaries from untrusted sources through VirusTotal before running them. Piping a Bash script from curl bypasses that. Furthermore, such Bash scripts, when they aren’t self-contained, make security checks more difficult than a self-contained archive, installer, or binary, even when downloading the script without immediate execution.
You could always curl the install script, and modify it to run the virus scan in between the build and install steps.
Nothing is stopping anyone from pointing their agent to that script to review and audit it before running it.
I don’t believe an agent can do that effectively without a sandbox to run the script in, if the script isn’t self-contained.
And everyone running a research agent on every download can’t be the solution. It’s much more effective to crowdsource a security database based on hashes. But for that, the downloads need to be self-contained.
It's `curl foo | sudo bash` that's the bigger objection. Running software usually shouldn't require root, and then the equivalence argument you make doesn't hold.
a script isn't getting hashed to see whether or not it's the one the website intended to serve you, for one.
what use is hashing every piece of software that goes thru the distros package manager just to throw caution to the wind at the layer above it?
w.r.t. "it's already from the same domain" , well most bash/z install scripts either invoke a package manager or they download and untar a package that has nothing to do with the host domain, anyway.
The real problem is that we just aren't using package managers. We should be using package managers. Package managers are really really good.
There is no argument. It's just people's reflex reactions.
The technical excuses they come up with (e.g. that the server can detect it and send different content) are just post-hoc justifications for their instinct.
Just ignore them.
Piping to bash is definitely worse because there is no plan mode in bash. Agents also normally don't execute anything transparently, at worst you'll see it doing something weird in the logs.
Yeah, I was surprised at first then had to dig through setup.py and setup.sh files to figure out.
And yet people will let AI agents run autonomously on their machines. I feel like we're reaching peak YOLO with security.
Why would I want to use a model that has no knowledge of Tiananmen Square or Winnie The Pooh?
Don't worry, soon the US will have models with no memory of Watergate or Jan 6.
I was expecting such a response. False equivalence. One is actual and one is speculative.
US models are absolutely censored as well
I see this kind of message a lot from people who haven't tried it. They have knowledge, just are trained to refuse to talk about it. Here's an abliterated Qwen3.8 Flash Next:
Here's the CoT: This is just misinformation, please stop spreading it. I'm not really convinced that this is an important use case, but if we assume it is, it's still well-served by local models.Ask Qwen about Xi looking like Winnie The Pooh.
The interesting thing here is that it's a model specialized fork of a generic inference engine that unlocks consumer hardware to run a bigger model with useable performance than it could before.
I don't understand what you gain by having your agent read and comment on HN for you? You already have met all karma thresholds to get full privilege and you don't seem to be a founder who's about to need name recognition to shill his next big thing.
Claude remove all punctuation so it looks like I wrote it myself. Yeah even the hyphens for adjectival compounds, fuck 'em, it's all punctuation so it's gotta go
botspeak
The emergence of model specific inference (for consumers) getting big performance wins is way more worth while to talk about than random comments on what people think about the qwen family of models. Even the resource management of Strata is less interesting. I think it's likely we'll start seeing more hand/llm crafted inference for different architectures.
dang did his bot read this as an instruction to comment again?