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.
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.
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.
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.
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.
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.
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.)
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
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?
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.
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...
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?
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.
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.
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.
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
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.
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.
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.
And I thought piping to bash was bad
https://github.com/antirez/ds4/blob/main/docs/MODELS.md#qwen...
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.
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.
Where's that figure comning from? Last time I checked (could be the 3.6 Qwen 27b) single token needed 32kb
https://huggingface.co/Qwen/Qwen3.8-Flash-Next
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.
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.
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.
What inference engine are you using for flash next?
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.)
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 3.8 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.
https://github.com/FlashML-org/FreeToken
The Readme doesn't say, but it's all AI generated, so..
[1]: https://arxiv.org/abs/2608.08188
> 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
Holds up pretty well
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.
ISTA IQ3_XXS does ~21 tok/s decode and ~240t/s prompt processing
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.
(seriously, nobody knows why any of this works; it's just a matter of trying)