Yeah, that's weird, seems it has later models, and earlier, but specifically not Pro 6000? Also, based on my experience, the given numbers seems to be at least one magnitude off, which seems like a lot, when I use the approx values for a Pro 6000 (96GB VRAM + 1792 GB/s)
1. I have an M3 Ultra with 256GB of memory, but the options list only goes up to 192GB. The M3 Ultra supports up to 512GB.
2. It'd be great if I could flip this around and choose a model, and then see the performance for all the different processors. Would help making buying decisions!
This is awesome, it would be great to cross reference some intelligence benchmarks so that I can understand the trade off between RAM consumption, token rate and how good the model is
"Available to userspace" is a much different thing than "available to every website that wants it, even in private mode".
I too was a little surprised by this. My browser (Vivladi) makes a big deal about how privacy-conscious they are, but apparently browser fingerprinting is not on their radar.
I run LibreWolf, which is configured to ask me before a site can use WebGL, which is commonly used for fingerprinting. I got the popup on this site, so I assume that's how they're doing it.
That's super handy, thanks for sharing the link. Way more useful than the web site this post is about, to be honest.
It looks like I can run more local LLMs than I thought, I'll have to give some of those a try. I have decent memory (96GB) but my M2 Max MBP is a few years old now and I figured it would be getting inadequate for the latest models. But llmfit thinks it's a really good fit for the vast majority of them. Interesting!
Your hardware can run a good range of local models, but keep an eye on quantization since 4-bit models trade off some accuracy, especially with longer context or tougher tasks. Thermal throttling is also an issue, since even Apple silicon can slow down when all cores are pushed for a while, so sustained performance might not match benchmark numbers.
This is great, I've been trying to figure this stuff out recently.
One thing I do wonder is what sort of solutions there are for running your own model, but using it from a different machine. I don't necessarily want to run the model on the machine I'm also working from.
Hugging Face can already do this for you (with much more up-to-date list of available models). Also LM Studio. However they don't attempt to estimate tok/sec, so that's a cool feature. However I don't really trust those numbers that much because it is not incorporating information about the CPU, etc. True GPU offload isn't often possible on consumer PC hardware. Also there are different quants available that make a big difference.
I've been trying to get speech to text to work with a reasonable vocabulary on pis for a while. It's tough. All the modern models just need more GPU than is available
I haven't tried on a raspberry pi, but on Intel it uses a little less than 1s of CPU time per second of audio. Using https://github.com/NVIDIA-NeMo/NeMo/blob/main/examples/asr/a... for chunked streaming inference, it takes 6 cores to process audio ~5x faster than realtime. I expect with all cores on a Pi 4 or 5, you'd probably be able to at least keep up with realtime.
(Batch inference, where you give it the whole audio file up front, is slightly more efficient, since chunked streaming inference is basically running batch inference on overlapping windows of audio.)
EDIT: there are also the multitalker-parakeet-streaming-0.6b-v1 and nemotron-speech-streaming-en-0.6b models, which have similar resource requirements but are built for true streaming inference instead of chunked inference. In my tests, these are slightly less accurate. In particular, they seem to completely omit any sentence at the beginning or end of a stream that was partially cut off.
This lacks a whole lot of mobile GPUs. It also does not understand that you can share CPU memory with the GPU, or perform various KV cache offloading strategies to work around memory limits.
It says I have an Arc 750 with 2 GB of shared RAM, because that's the GPU that renders my browser...but I actually have an RTX1000 Ada with 6 GB of GDDR6. It's kind of like an RTX 4050 (not listed in the dropdowns) with lower thermal limits. I also have 64 GB of LPDDR5 main memory.
It works - Qwen3 Coder Next, Devstral Small, Qwen3.5 4B, and others can run locally on my laptop in near real-time. They're not quite as good as the latest models, and I've tried some bigger ones (up to 24GB, it produces tokens about half as fast as I can type...which is disappointingly slow) that are slower but smarter.
The RAM/VRAM cutoff matters more than the parameter count alone. A 13B model in Q4_K_M quantization fits in 8GB VRAM with reasonable throughput, but the same model in fp16 needs 26GB. Most calculators treat quantization as a footnote when it is actually the primary variable. The question is not "can I run 13B" but "what quantization level gives acceptable quality at my hardware ceiling".
This is the right framing. I'd add that quantization is only the first dimension -- the second is what you build around the model. A Q4_K_M 14B model running raw inference vs. the same model with structured constraint extraction, diverse candidate sampling, and iterative self-repair are essentially different systems despite identical VRAM footprint.
The real question isn't "what quantization gives acceptable quality at my hardware ceiling" -- it's "what inference pipeline gives acceptable quality at my hardware ceiling." A single-shot Q4_K_M 14B will disappoint you. The same model generating 3 candidates, scoring them with self-embeddings, and self-repairing failures will surprise you. Same GPU, same VRAM, just smarter infrastructure.
Having the rating of how well the model will run for you is cool. I miss to also have some rating of the model capabilities (even if this is tricky). There are way too many to choose. And just looking at the parameter number or the used memory is not always a good indication of actual performance.
Is there a reliable guide somewhere to setting up local AI for coding (please don’t say ‘just Google it’ - that just results in a morass of AI slop/SEO pages with out of date, non-self-consistent, incorrect or impossible instructions).
I’d like to be able to use a local model (which one?) to power Copilot in vscode, and run coding agent(s) (not general purpose OpenClaw-like agents) on my M2 MacBook. I know it’ll be slow.
I suspect this is actually fairly easy to set up - if you know how.
You're probably not going to get anything working well as an agent on an M2 MacBook, but smaller models do surprisingly well for focused autocomplete. Maybe the Qwen3.5 9B model would run decently on your system?
For LM Studio under server settings you can start a local server that has an OpenAI-compatible API. You'd need to point Copilot to that. I don't use Copilot so not sure of the exact steps there
I tried the Zed editor and it picked up Ollama with almost no fiddling, so that has allowed me to run Qwen3.5:9B just by tweaking the ollama settings (which had a few dumb defaults, I thought, like assuming I wanted to run 3 LLMs in parallel, initially disabling Flash Attention, and having a very short context window...).
Having a second pair of "eyes" to read a log error and dig into relevant code is super handy for getting ideas flowing.
This doesn't look accurate to me. I have an RX9070 and I've been messing around with Qwen 3.5 35B-A3B. According to this site I can't even run it, yet I'm getting 32tok/s ^.-
if you do, would you still want to collect data in a single pane of glass? see my open source repo for aggregating harness data from multiple machine learning model harnesses & models into a single place to discover what you are working on & spending time & money. there is plans for a scrobble feature like last.fm but for agent research & code development & execution.
thanks, I'll check for comments, feel free to fork but if you want to contribute you'll have to find me off of github, I develop privately on my own self hosted gitlab server. good luck & God bless.
Does it make any sense? I tried few models at 128GB and it's all pretty much rubbish. Yes they do give coherent answers, sometimes they are even correct, but most of the time it is just plain wrong. I find it massive waste of time.
For some reason it doesn't react to changing the RAM amount in the combo box at the top. If I open this on my Ryzen AI Max 395+ with 32 GB of unified memory, it thinks nothing will fit because I've set it up to reserve 512MB of RAM for the GPU.
Yeah, this site is iffy at best. I didn't even see Strix Halo on the list, but I selected 128GB and bumped up the memory bandwidth. It says gpt-oss-120b "barely runs" at ~2 t/s.
In reality, gpt-oss-120b fits great on the machine with plenty of room to spare and easily runs inference north of 50 t/s depending on context.
My Mac mini rocks qwen2.5 14b at a lightning fast 11/tokens a second. Which is actually good enough for the long term data processing I make it spend all day doing. It doesn’t lock up the machine or prevent its primary purpose as webserver from being fulfilled.
i think the perplexity is more important than tokens per second. tokens per second is relatively useless in my opinion. there is nothing worse than getting bad results returned to you very quickly and confidently.
ive been working with quite a few open weight models for the last year and especially for things like images, models from 6 months would return garbage data quickly, but these days qwen 3.5 is incredible, even the 9b model.
This seems to be estimating based on memory bandwidth / size of model, which is a really good estimate for dense models, but MoE models like GPT-OSS-20b don't involve the entire model for every token, so they can produce more tokens/second on the same hardware. GPT-OSS-20B has 3.6B active parameters, so it should perform similarly to a 3-4B dense model, while requiring enough VRAM to fit the whole 20B model.
(In terms of intelligence, they tend to score similarly to a dense model that's as big as the geometric mean of the full model size and the active parameters, i.e. for GPT-OSS-20B, it's roughly as smart as a sqrt(20b*3.6b) ≈ 8.5b dense model, but produces tokens 2x faster.)
Yeah, I looked up some models I have actually run locally on my Strix Halo laptop, and its saying I should have much lower performance than I actually have on models I've tested.
For MoE models, it should be using the active parameters in memory bandwidth computation, not the total parameters.
While your remark is valid, there's two small inaccuracies here:
> GPT-OSS-20B has 3.6B active parameters, so it should perform similarly to a 3-4B dense model, while requiring enough VRAM to fit the whole 20B model.
First, the token generation speed is going to be comparable, but not the prefil speed (context processing is going to be much slower on a big MoE than on a small dense model).
Second, without speculative decoding, it is correct to say that a small dense model and a bigger MoE with the same amount of active parameters are going to be roughly as fast. But if you use a small dense model you will see token generation performance improvements with speculative decoding (up to x3 the speed), whereas you probably won't gain much from speculative decoding on a MoE model (because two consecutive tokens won't trigger the same “experts”, so you'd need to load more weight to the compute units, using more bandwidth).
So, this is all true, but this calculation isn't that nuanced. It's trying to get you into a ballpark range, and based on my usage on my real hardware (if I put in my specs, since it's not in their hardware list), the results are fairly close to my real experience if I compensate for the issue where it's calculating based on total params instead of active.
So by doing so, this calculator is telling you that you should be running entirely dense models, and sparse MoE models that maybe both faster and perform better are not recommended.
> A Mixture of Experts model splits its parameters into groups called "experts." On each token, only a few experts are active — for example, Mixtral 8x7B has 46.7B total parameters but only activates ~12.9B per token. This means you get the quality of a larger model with the speed of a smaller one. The tradeoff: the full model still needs to fit in memory, even though only part of it runs at inference time.
> A dense model activates all its parameters for every token — what you see is what you get. A MoE model has more total parameters but only uses a subset per token. Dense models are simpler and more predictable in terms of memory/speed. MoE models can punch above their weight in quality but need more VRAM than their active parameter count suggests.
It discusses it, and they have data showing that they know the number of active parameters on an MoE model, but they don't seem to use that in their calculation. It gives me answers far lower than my real-world usage on my setup; its calculation lines up fairly well for if I were trying to run a dense model of that size. Or, if I increase my memory bandwidth in the calculator by a factor of 10 or so which is the ratio between active and total parameters in the model, I get results that are much closer to real world usage.
I'm guessing this is also calculating based on the full context size that the model supports but depending on your use case it will be misleading. Even on a small consumer card with Qwen 3 30B-A3B you probably don't need 128K context depending on what you're doing so a smaller context and some tensor overrides will help. llama.cpp's llama-fit-params is helpful in those cases.
tbh i stopped caring about "can i run X locally" a while ago. for anything where quality matters (scripting, code, complex reasoning) the local models are just not there yet compared to API. where local shines is specific narrow tasks - TTS, embeddings, whisper for STT, stuff like that. trying to run a 70b model at 3 tok/s on your gaming GPU when you could just hit an API for like $0.002/req feels like a weird flex IMO
The "local models aren't there yet" take was accurate 12 months ago, but things have moved fast. A frozen Qwen3-14B at Q4_K_M on a single 16GB consumer GPU can clear 70%+ on LiveCodeBench pass@1 if you wrap it in the right inference pipeline -- structured generation, best-of-k candidate sampling, self-verified iterative repair. That puts it in the ballpark of Claude 4 Sonnet's single-shot score.
The insight most people miss is that "running locally" doesn't have to mean "single-shot raw inference and hope for the best." The model is the engine, not the car. You can build constraint extraction, budget-controlled thinking, and self-repair loops around a frozen model and get results that would have seemed impossible at that parameter count a year ago. Cost works out to fractions of a cent per task in electricity.
For narrow tasks like embeddings and TTS, sure, local has always been fine. But for coding and reasoning, the gap has closed dramatically -- you just have to stop treating local inference as "discount API" and start treating it as a compute substrate you control.
When running models on my phone - either through the web browser or via an app - is there any chance it uses the phone's NPU, or will these be GPU only?
I don't really understand how the interface to the NPU chip looks from the perspective of a non-system caller, if it exists at all. This is a Samsung device but I am wondering about the general principle.
It would be great if something like this was built into ollama, so you could easily list available models based on your current hardware setup, from the CLI.
The latest level of abstraction! You just release your ideas half baked in some internet connected box and wake up with products! Yahoo! Onwards into the Gestell!
You can still run larger MoE models using expert weight off-loading to the CPU for token generation. They are by and large useable, I get ~50 toks/second on a kimi linear 48B (3B active) model on a potato PC + a 3090
If anyone hasn't tried Qwen3.5 on Apple Silicon, I highly suggest you to! Claude level performance on local hardware. If the Qwen team didn't get fired, I would be bullish on Local LLM.
Open with multiple browsers (safari vs chrome) to get more "accurate + glanceable" rankings.
Its using WebGPU as a proxy to estimate system resource.
Chrome tends to leverage as much resources (Compute + Memory) as the OS makes available. Safari tends to be more efficient.
Maybe this was obvious to everyone else. But its worth re-iterating for those of us skimmers of HN :)
I have spent a HUGE amount of time the last two years experimenting with local models.
A few lessons learned:
1. small models like the new qwen3.5:9b can be fantastic for local tool use, information extraction, and many other embedded applications.
2. For coding tools, just use Google Antigravity and gemini-cli, or, Anthropic Claude, or...
Now to be clear, I have spent perhaps 100 hours in the last year configuring local models for coding using Emacs, Claude Code (configured for local), etc. However, I am retired and this time was a lot of fun for me: lot's of efforts trying to maximize local only results. I don't recommend it for others.
I do recommend getting very good at using embedded local models in small practical applications. Sweet spot.
What kind of hardware did you use? I suppose that a 8GB gaming GPU and a Mac Pro with 512 GB unified RAM give quite different results, both formally being local.
I've been really interested in the difference between 3.5 9b and 14b for information extraction. Is there a discernible difference in quality of capability?
I'd love to know how you fit smaller models into your workflow. I have an M4 Macbook Pro w/ 128GB RAM and while I have toyed with some models via ollama, I haven't really found a nice workflow for them yet.
It really depends on the tasks you have to perform. I am using specialized OCR models running locally to extract page layout information and text from scanned legal documents. The quality isn't perfect, but it is really good compared to desktop/server OCR software that I formerly used that cost hundreds or thousands of dollars for a license. If you have similar needs and the time to try just one model, start with GLM-OCR.
If you want a general knowledge model for answering questions or a coding agent, nothing you can run on your MacBook will come close to the frontier models. It's going to be an exercise in frustration if you try to use local models that way. But there are a lot of useful applications for local-sized models when it comes to describing and categorizing unstructured data.
I use Raycast and connect it to LM Studio to run text clean up and summaries often. The models are small enough I keep them in memory more often than not
Most workstation class laptops (i.e. Lenovo P-series, Dell Precision) have 4 DIMM slots and you can get them with 256 GB (at least, before the current RAM shortages).
There's also the Ryzen AI Max+ 395 that has 128GB unified in laptop form factor.
Only Apple has the unique dynamic allocation though.
Yep, I have a 13" gaming tablet with the 128 GB AMD Strix Halo chip (Ryzen AI Max+ 395, what a name). Asus ROG Flow Z13. It's a beast; the performance is totally disproportionate to its size & form factor.
I'm not sure what exactly you're referring to with "Only Apple has the unique dynamic allocation though." On Strix Halo you set the fixed VRAM size to 512 MB in the BIOS, and you set a few Linux kernel params that enable dynamic allocation to whatever limit you want (I'm using 110 GB max at the moment). LLMs can use up to that much when loaded, but it's shared fully dynamically with regular RAM and is instantly available for regular system use when you unload the LLM.
Is it correct that there's zero improvement in performance between M4 (+Pro/Max) and M5 (+Pro/Max) the data looks identical. Also the memory does not seem to improve performance on larger models when I thought it would have?
Love the idea though!
EDIT: Okay the whole thing is nonsense and just some rough guesswork or asking an LLM to estimate the values. You should have real data (I'm sure people here can help) and put ESTIMATE next to any of the combinations you are guessing.
The M4 Ultra doesn't exist and there is more credible rumors for an M5 Ultra. I wouldn't put a projection like that without highlighting that this processor doesn't exist yet.
This is kind of bogus since some of the S and A tier models are pretty useless for reasoning or tool calls and can’t run with any sizable system prompt… it seems to be solely based on tokens per second?
This does not seem accurate based on my recently received M5 Max 128GB MBP. I think there's some estimates/guesswork involved, and it's also discounting that you can move the memory divider on Unified Memory devices like Apple Silicon and AMD AI Max 395+.
This (+ llmfit) are great attempts, but I've been generally frustrated by how it feels so hard to find any sort of guidance about what I would expect to be the most straightforward/common question:
"What is the highest-quality model that I can run on my hardware, with tok/s greater than <x>, and context limit greater than <y>"
(My personal approach has just devolved into guess-and-check, which is time consuming.) When using TFA/llmfit, I am immediately skeptical because I already know that Qwen 3.5 27B Q6 @ 100k context works great on my machine, but it's buried behind relatively obsolete suggestions like the Qwen 2.5 series.
I'm assuming this is because the tok/s is much higher, but I don't really get much marginal utility out of tok/s speeds beyond ~50 t/s, and there's no way to sort results by quality.
It’s a hard problem. I’ve been working on it for the better part of a year.
Well, granted my project is trying to do this in a way that works across multiple devices and supports multiple models to find the best “quality” and the best allocation. And this puts an exponential over the project.
But “quality” is the hard part. In this case I’m just choosing the largest quants.
LLMs are just special purpose calculators, as opposed to normal calculators which just do numbers and MUST be accurate. There aren't very good ways of knowing what you want because the people making the models can't read your mind and have different goals
Great tool for local inference. The flip side question is always 'should I run it locally or use a cloud API?' The answer depends heavily on volume and current vendor pricing. Cloud inference costs have been surprisingly volatile lately — we tracked 30 price changes across 615 models just this week.
Tools like this are crucial for the local AI movement. What I've found in practice is that the 7-8B parameter models with Q4_K_M quantization hit a sweet spot for most developer machines, giving you 90%+ of the capability at a fraction of the memory footprint. The bigger unlock here isn't just cost savings though, it's data sovereignty. When you can run inference without your prompts leaving your machine, you can actually use LLMs for sensitive code reviews, proprietary data analysis, and internal tooling that you'd never trust to a cloud API. Would love to see this tool also flag which models have good tool-calling support since that's increasingly what separates "neat demo" from "production-ready."
Great tool for local inference. The flip side question is always 'should I run it locally or use a cloud API?' The answer depends heavily on volume and current vendor pricing. Cloud inference costs have been surprisingly volatile lately. We tracked 30 price changes across 615 models just this week.
Tools like this are crucial for the local AI movement. What I've found in practice is that the 7-8B parameter models with Q4_K_M quantization hit a sweet spot for most developer machines, giving you 90%+ of the capability at a fraction of the memory footprint. The bigger unlock here isn't just cost savings though, it's data sovereignty. When you can run inference without your prompts leaving your machine, you can actually use LLMs for sensitive code reviews, proprietary data analysis, and internal tooling that you'd never trust to a cloud API. Would love to see this tool also flag which models have good tool-calling support since that's increasingly what separates "neat demo" from "production-ready."
Sorry if already been answered, but will there be a metric for latency aka time to first token?
Since I considered buying M3 Ultra and feel like it the most often discussed regarding using Apple hardware for runninh local LLMs. Where speed might be okay, but prompt processing can take ages.
I'm not usually one to whine, but agreed; additionally, add contrast to the modifiers (e.g. processor select). First thing I did when I visited was scale the website to 150%
Super impressive comparisons, and correlates with my perception having three seperate generations of GPU (from your list pulldown). Thanks for including the "old AMD" Polaris chipsets, which are actually still much faster than lower-spec Apple silicon. I have Ollama3.1 on a VEGA64 and it really is twice as fast as an M2Pro...
----
For anybody that thinks installing a local LLM is complicated: it's not (so long as you have more than one computer, don't tinker on your primary workhorse). I am a blue collar electrician (admittedly: geeky); no more difficult than installing linux.
From my personal testing, running various agentic tasks with a bunch of tool calls on an M4 Max 128GB, I've found that running quantized versions of larger models to produce the best results which this site completely ignores.
Currently, Nemotron 3 Super using Unsloth's UD Q4_K_XL quant is running nearly everything I do locally (replacing Qwen3.5 122b)
I have amd 9700 and it is not listed while it is great llm hardware because it has 32Gb for a reasonable price. I tried doing "custom" but it didn't seem to work.
John23832 | 4 hours ago
schaefer | 4 hours ago
embedding-shape | 3 hours ago
sxates | 4 hours ago
A couple suggestions:
1. I have an M3 Ultra with 256GB of memory, but the options list only goes up to 192GB. The M3 Ultra supports up to 512GB. 2. It'd be great if I could flip this around and choose a model, and then see the performance for all the different processors. Would help making buying decisions!
utopcell | 57 minutes ago
ProllyInfamous | 49 minutes ago
GrayShade | 4 hours ago
Aurornis | 2 hours ago
GrayShade | an hour ago
$ ./llama-cli unsloth_Qwen3.5-35B-A3B-GGUF_Qwen3.5-35B-A3B-UD-Q4_K_XL.gguf -c 65536 -p "Hello"
[snip 73 lines]
[ Prompt: 86,6 t/s | Generation: 34,8 t/s ]
$ ./llama-cli unsloth_Qwen3.5-35B-A3B-GGUF_Qwen3.5-35B-A3B-UD-Q4_K_XL.gguf -c 262144 -p "Hello"
[snip 128 lines]
[ Prompt: 78,3 t/s | Generation: 30,9 t/s ]
I suspect the ROCm build will be faster, but it doesn't work out of the box for me.
phelm | 3 hours ago
S4phyre | 3 hours ago
adithyassekhar | 3 hours ago
Not sure if it still works.
twampss | 3 hours ago
https://github.com/AlexsJones/llmfit
deanc | 3 hours ago
dgrin91 | 3 hours ago
dekhn | 3 hours ago
bityard | 2 hours ago
I too was a little surprised by this. My browser (Vivladi) makes a big deal about how privacy-conscious they are, but apparently browser fingerprinting is not on their radar.
swiftcoder | 2 hours ago
dekhn | 2 hours ago
rithdmc | 2 hours ago
> Estimates based on browser APIs. Actual specs may vary
spudlyo | 2 hours ago
johnisgood | an hour ago
You can check out here how it does that: https://github.com/AlexsJones/llmfit/blob/main/llmfit-core/s...
To detect NVIDIA GPUs, for example: https://github.com/AlexsJones/llmfit/blob/main/llmfit-core/s...
In this case it just runs the command "nvidia-smi".
Note: llmfit is not web-based.
Someone1234 | 15 minutes ago
- If you already HAVE a computer and are looking for models: LLMFit
- If you are looking to BUY a computer/hardware, and want to compare/contrast for local LLM usage: This
You cannot exactly run LLMFit on hardware you don't have.
rootusrootus | 2 hours ago
It looks like I can run more local LLMs than I thought, I'll have to give some of those a try. I have decent memory (96GB) but my M2 Max MBP is a few years old now and I figured it would be getting inadequate for the latest models. But llmfit thinks it's a really good fit for the vast majority of them. Interesting!
hrmtst93837 | 25 minutes ago
mrdependable | 3 hours ago
One thing I do wonder is what sort of solutions there are for running your own model, but using it from a different machine. I don't necessarily want to run the model on the machine I'm also working from.
cortesoft | 3 hours ago
You can also use the kubernetes operator to run them on a cluster: https://ollama-operator.ayaka.io/pages/en/
rebolek | 2 hours ago
g_br_l | 3 hours ago
vova_hn2 | 3 hours ago
charcircuit | 3 hours ago
debatem1 | 3 hours ago
Just FYI.
metalliqaz | 3 hours ago
havaloc | 3 hours ago
arjie | 3 hours ago
ge96 | 3 hours ago
I also want to run vision like Yocto and basic LLM with TTS/STT
boutell | 2 hours ago
ge96 | 2 hours ago
For wakewords I have used pico rhino voice
I want to use these I2S breakout mics
meatmanek | 2 hours ago
I haven't tried on a raspberry pi, but on Intel it uses a little less than 1s of CPU time per second of audio. Using https://github.com/NVIDIA-NeMo/NeMo/blob/main/examples/asr/a... for chunked streaming inference, it takes 6 cores to process audio ~5x faster than realtime. I expect with all cores on a Pi 4 or 5, you'd probably be able to at least keep up with realtime.
(Batch inference, where you give it the whole audio file up front, is slightly more efficient, since chunked streaming inference is basically running batch inference on overlapping windows of audio.)
EDIT: there are also the multitalker-parakeet-streaming-0.6b-v1 and nemotron-speech-streaming-en-0.6b models, which have similar resource requirements but are built for true streaming inference instead of chunked inference. In my tests, these are slightly less accurate. In particular, they seem to completely omit any sentence at the beginning or end of a stream that was partially cut off.
LeifCarrotson | 3 hours ago
It says I have an Arc 750 with 2 GB of shared RAM, because that's the GPU that renders my browser...but I actually have an RTX1000 Ada with 6 GB of GDDR6. It's kind of like an RTX 4050 (not listed in the dropdowns) with lower thermal limits. I also have 64 GB of LPDDR5 main memory.
It works - Qwen3 Coder Next, Devstral Small, Qwen3.5 4B, and others can run locally on my laptop in near real-time. They're not quite as good as the latest models, and I've tried some bigger ones (up to 24GB, it produces tokens about half as fast as I can type...which is disappointingly slow) that are slower but smarter.
But I don't run out of tokens.
Felixbot | 3 hours ago
itigges22 | an hour ago
The real question isn't "what quantization gives acceptable quality at my hardware ceiling" -- it's "what inference pipeline gives acceptable quality at my hardware ceiling." A single-shot Q4_K_M 14B will disappoint you. The same model generating 3 candidates, scoring them with self-embeddings, and self-repairing failures will surprise you. Same GPU, same VRAM, just smarter infrastructure.
sshagent | 3 hours ago
carra | 3 hours ago
jrmg | 3 hours ago
I’d like to be able to use a local model (which one?) to power Copilot in vscode, and run coding agent(s) (not general purpose OpenClaw-like agents) on my M2 MacBook. I know it’ll be slow.
I suspect this is actually fairly easy to set up - if you know how.
AstroBen | 2 hours ago
You're probably not going to get anything working well as an agent on an M2 MacBook, but smaller models do surprisingly well for focused autocomplete. Maybe the Qwen3.5 9B model would run decently on your system?
jrmg | 2 hours ago
brcmthrowaway | 2 hours ago
Try this article https://advanced-stack.com/fields-notes/qwen35-opencode-lm-s...
I'm looking for an alternative to OpenCode though, I can barely see the UI.
AstroBen | 2 hours ago
AstroBen | 2 hours ago
For LM Studio under server settings you can start a local server that has an OpenAI-compatible API. You'd need to point Copilot to that. I don't use Copilot so not sure of the exact steps there
NortySpock | 2 hours ago
Having a second pair of "eyes" to read a log error and dig into relevant code is super handy for getting ideas flowing.
chatmasta | an hour ago
AstroBen | 2 hours ago
misnome | 2 hours ago
unfirehose | 2 hours ago
https://github.com/russellballestrini/unfirehose-nextjs-logg...
thanks, I'll check for comments, feel free to fork but if you want to contribute you'll have to find me off of github, I develop privately on my own self hosted gitlab server. good luck & God bless.
varispeed | 2 hours ago
boutell | 2 hours ago
orthoxerox | 2 hours ago
bityard | 2 hours ago
In reality, gpt-oss-120b fits great on the machine with plenty of room to spare and easily runs inference north of 50 t/s depending on context.
kylehotchkiss | 2 hours ago
freediddy | 2 hours ago
ive been working with quite a few open weight models for the last year and especially for things like images, models from 6 months would return garbage data quickly, but these days qwen 3.5 is incredible, even the 9b model.
sroussey | 2 hours ago
But yes, if there is a choice I want quality over speed. At same quality, I definitely want speed.
meatmanek | 2 hours ago
(In terms of intelligence, they tend to score similarly to a dense model that's as big as the geometric mean of the full model size and the active parameters, i.e. for GPT-OSS-20B, it's roughly as smart as a sqrt(20b*3.6b) ≈ 8.5b dense model, but produces tokens 2x faster.)
lambda | 2 hours ago
For MoE models, it should be using the active parameters in memory bandwidth computation, not the total parameters.
littlestymaar | 2 hours ago
> GPT-OSS-20B has 3.6B active parameters, so it should perform similarly to a 3-4B dense model, while requiring enough VRAM to fit the whole 20B model.
First, the token generation speed is going to be comparable, but not the prefil speed (context processing is going to be much slower on a big MoE than on a small dense model).
Second, without speculative decoding, it is correct to say that a small dense model and a bigger MoE with the same amount of active parameters are going to be roughly as fast. But if you use a small dense model you will see token generation performance improvements with speculative decoding (up to x3 the speed), whereas you probably won't gain much from speculative decoding on a MoE model (because two consecutive tokens won't trigger the same “experts”, so you'd need to load more weight to the compute units, using more bandwidth).
lambda | 59 minutes ago
So by doing so, this calculator is telling you that you should be running entirely dense models, and sparse MoE models that maybe both faster and perform better are not recommended.
littlestymaar | 56 minutes ago
But since this is a tech forum, I assumed some people would be interested by the correction on the details that were wrong.
pbronez | 2 hours ago
> A Mixture of Experts model splits its parameters into groups called "experts." On each token, only a few experts are active — for example, Mixtral 8x7B has 46.7B total parameters but only activates ~12.9B per token. This means you get the quality of a larger model with the speed of a smaller one. The tradeoff: the full model still needs to fit in memory, even though only part of it runs at inference time.
> A dense model activates all its parameters for every token — what you see is what you get. A MoE model has more total parameters but only uses a subset per token. Dense models are simpler and more predictable in terms of memory/speed. MoE models can punch above their weight in quality but need more VRAM than their active parameter count suggests.
https://www.canirun.ai/docs
lambda | an hour ago
tommy_axle | an hour ago
nilslindemann | 2 hours ago
2. Add a 150% size bonus to your site.
Otherwise, cool site, bookmarked.
amelius | 2 hours ago
amelius | 2 hours ago
vikramkr | 2 hours ago
relaxing | 2 hours ago
swiftcoder | 2 hours ago
tcbrah | 2 hours ago
itigges22 | an hour ago
The insight most people miss is that "running locally" doesn't have to mean "single-shot raw inference and hope for the best." The model is the engine, not the car. You can build constraint extraction, budget-controlled thinking, and self-repair loops around a frozen model and get results that would have seemed impossible at that parameter count a year ago. Cost works out to fractions of a cent per task in electricity.
For narrow tasks like embeddings and TTS, sure, local has always been fine. But for coding and reasoning, the gap has closed dramatically -- you just have to stop treating local inference as "discount API" and start treating it as a compute substrate you control.
hatthew | 30 minutes ago
sdingi | 2 hours ago
I don't really understand how the interface to the NPU chip looks from the perspective of a non-system caller, if it exists at all. This is a Samsung device but I am wondering about the general principle.
amelius | 2 hours ago
rootusrootus | 2 hours ago
Quick, someone go vibe code that.
dugidugout | 18 minutes ago
am17an | 2 hours ago
brcmthrowaway | 2 hours ago
golem14 | 2 hours ago
The website says that code export is not working yet.
That’s a very strange way to advertise yourself.
cafed00d | an hour ago
Its using WebGPU as a proxy to estimate system resource. Chrome tends to leverage as much resources (Compute + Memory) as the OS makes available. Safari tends to be more efficient.
Maybe this was obvious to everyone else. But its worth re-iterating for those of us skimmers of HN :)
ryandrake | an hour ago
mark_l_watson | an hour ago
A few lessons learned:
1. small models like the new qwen3.5:9b can be fantastic for local tool use, information extraction, and many other embedded applications.
2. For coding tools, just use Google Antigravity and gemini-cli, or, Anthropic Claude, or...
Now to be clear, I have spent perhaps 100 hours in the last year configuring local models for coding using Emacs, Claude Code (configured for local), etc. However, I am retired and this time was a lot of fun for me: lot's of efforts trying to maximize local only results. I don't recommend it for others.
I do recommend getting very good at using embedded local models in small practical applications. Sweet spot.
nine_k | an hour ago
fzzzy | 29 minutes ago
nine_k | 13 minutes ago
manmal | an hour ago
kylehotchkiss | an hour ago
johnmaguire | an hour ago
philipkglass | an hour ago
If you want a general knowledge model for answering questions or a coding agent, nothing you can run on your MacBook will come close to the frontier models. It's going to be an exercise in frustration if you try to use local models that way. But there are a lot of useful applications for local-sized models when it comes to describing and categorizing unstructured data.
saltwounds | 58 minutes ago
Bluecobra | 39 minutes ago
AzN1337c0d3r | 34 minutes ago
There's also the Ryzen AI Max+ 395 that has 128GB unified in laptop form factor.
Only Apple has the unique dynamic allocation though.
the_pwner224 | 19 minutes ago
I'm not sure what exactly you're referring to with "Only Apple has the unique dynamic allocation though." On Strix Halo you set the fixed VRAM size to 512 MB in the BIOS, and you set a few Linux kernel params that enable dynamic allocation to whatever limit you want (I'm using 110 GB max at the moment). LLMs can use up to that much when loaded, but it's shared fully dynamically with regular RAM and is instantly available for regular system use when you unload the LLM.
andy_ppp | an hour ago
Love the idea though!
EDIT: Okay the whole thing is nonsense and just some rough guesswork or asking an LLM to estimate the values. You should have real data (I'm sure people here can help) and put ESTIMATE next to any of the combinations you are guessing.
GeekyBear | an hour ago
Preliminary testing did not come to that conclusion.
> Apple’s New M5 Max Changes the Local AI Story
https://www.youtube.com/watch?v=XGe7ldwFLSE
lostmsu | 52 minutes ago
For the lazy: that's less then 3x: 1.8 * 3 = 5.4
mkagenius | an hour ago
zitterbewegung | an hour ago
rcarmo | an hour ago
polyterative | an hour ago
tristor | an hour ago
bheadmaster | an hour ago
tencentshill | an hour ago
mopierotti | an hour ago
"What is the highest-quality model that I can run on my hardware, with tok/s greater than <x>, and context limit greater than <y>"
(My personal approach has just devolved into guess-and-check, which is time consuming.) When using TFA/llmfit, I am immediately skeptical because I already know that Qwen 3.5 27B Q6 @ 100k context works great on my machine, but it's buried behind relatively obsolete suggestions like the Qwen 2.5 series.
I'm assuming this is because the tok/s is much higher, but I don't really get much marginal utility out of tok/s speeds beyond ~50 t/s, and there's no way to sort results by quality.
J_Shelby_J | an hour ago
Well, granted my project is trying to do this in a way that works across multiple devices and supports multiple models to find the best “quality” and the best allocation. And this puts an exponential over the project.
But “quality” is the hard part. In this case I’m just choosing the largest quants.
downrightmike | 40 minutes ago
A7OM | an hour ago
JulianPembroke | an hour ago
A7OM | an hour ago
JulianPembroke | an hour ago
SXX | an hour ago
Since I considered buying M3 Ultra and feel like it the most often discussed regarding using Apple hardware for runninh local LLMs. Where speed might be okay, but prompt processing can take ages.
teaearlgraycold | an hour ago
tkfoss | an hour ago
anigbrowl | an hour ago
lagrange77 | an hour ago
mmaunder | 51 minutes ago
ProllyInfamous | 41 minutes ago
Super impressive comparisons, and correlates with my perception having three seperate generations of GPU (from your list pulldown). Thanks for including the "old AMD" Polaris chipsets, which are actually still much faster than lower-spec Apple silicon. I have Ollama3.1 on a VEGA64 and it really is twice as fast as an M2Pro...
----
For anybody that thinks installing a local LLM is complicated: it's not (so long as you have more than one computer, don't tinker on your primary workhorse). I am a blue collar electrician (admittedly: geeky); no more difficult than installing linux.
reactordev | 48 minutes ago
This isn’t nearly complete.
azmenak | 47 minutes ago
Currently, Nemotron 3 Super using Unsloth's UD Q4_K_XL quant is running nearly everything I do locally (replacing Qwen3.5 122b)
bearjaws | 37 minutes ago
kuon | 31 minutes ago
The tool is very nice though.
ipunchghosts | 28 minutes ago