Nice setup; but, for simple tasks or questions, AI is currently free? And it will probably stay free, as I don't see Google starting to charge for using AI on its search engine? So costs can't be a motivation for running small models locally?
For more complex or important tasks, costs, autonomy and privacy matter, but then so does performance/quality.
So I'm not completely convinced it's really worth it; but it's tempting!
I'm the author - hello! I talk about it in the blog post - knowing what's being run, knowing where it's being run, and not having anyone else control it.
I figure most free AI is free as in free electricity in the coffee shop. You're welcome to use it for small reasonable loads, but try to build anything off of it and you'll soon find yourself barred from the establishment.
And that's probably good, otherwise the free ai would just be unavailable for everyone else
Maybe, "Free as in free WiFi?" Like WiFi, the models you can use for free online aren't the highest quality, and can be pulled any time.
The models used in TFA are halfway in between the traditional "free as in beer" software. Open weight means once you download it, it continues to work forever; and you can also do your own RL on them; but you can't really see what went into their training, nor train a new one yourself from scratch.
No mention of the performance of the models? I'm able to load a bunch of different models on my little mini-PC with 16GB RAM, but the performance is terrible. I always wonder what performance people are getting with local models that they find is acceptable?
i have a 512gb ram m3 ultra mac studio setup with a gas city that runs one of my companies. today was the first time ever that a local model (GLM5.3 8-bit) was able to match fable5 in our tests.
GLM-5.3-Flash at true 8-bit: 341 GB on disk, 328 GB resident, 288 experts across 46 layers, loads in 65 seconds.
• 18.7 tokens/s generation, 35 tokens/s prompt, on a desk, on a $0 per-token bill.
• Runs beside our whole agent city on one box with ~130 GB to spare.
• Review test: caught 6 of 6 planted P1 defects, zero false positives, same score as the frontier model we pay for.
• CRM test: 11 of 11 required records extracted, zero wrong writes, 45 minutes, first local model to clear the bar.
• Serving a 131k-token window today; the model itself supports 1,048,576. Widened to 4 concurrent slots and still have 50gb+ of excess ram.
granted my cto still isn't moving all of our inference to glm5.3 but we've identified 40%+ that is currently handled by fable that we're routing locally instead and will do concurrent requests to verify/compare responses for a while.
You still have electricity and capital investment. Envelope math suggests cheap electricity is costing you something like $0.50/mtok and the opportunity cost on the capital tied up and lost in the unit purchase and resale is going to cost you something like $2/mtok at 100% utilization (so, frontier model prices or higher at real utilization), and you don't benefit from any elasticity.
Hosted GLM 5.3 flash is like $0.15/mtok in $0.50/mtok out
Time to completion also must be considered. If I have to wait around for hours for a prompt to complete locally and I’ll need to iterate quickly, I’m better off hosted than local. If it’s “free” and slow it may just not be worth it.
I run a similar setup to the one he described on similar hardware. I run bifrost and llama swap though (tailscale rocks). My local model usage is for some out of band batch processing one of my personal apps uses. Basically a personalized recommender for media, it curates stuff for me based on a database i've compiled over years, so non-interactive. For that use case, I don't really care that it might take a few minutes to run. It's free. The machine is just sitting there anyway. I have tried using qwen-coder and opencode on my M5 Max 128gb and compared to claude code it's painful. I did setup a workflow where claude plans, qwen executes (unattended overnight, again b/c it's slow) and then claude reviews. I benchmarked this several times and I ended up using MORE tokens with claude because it had to 'fix' all the qwen issues. While the code it produced was 'good enough' the fixes were worth it so I just stick to coding task using API models (codex and claude).
It isn't, the cost is included in your electricity bill, not even talking about the cost of your time to set it up. It's very possible that it costs you more than a cloud mode would, you just don't want to calculate it properly.
> It's very possible that it costs you more than a cloud mode would
...which is almost always true in a single request/reply mode and never true in batch mode. Single request usually 2x-3x more expensive than cloud and batch mode 2x-3x cheaper. Now, for narrow tasks, a finetuned tiny 8b model would dramatically outperform SOTA frontiers for a fraction of price, esp. on energy efficient hardware like Apple.
Local is never cheaper than cloud because they can do batch inference, and that means you load model weights once to produce 128 tokens on 128 sessions in parallel not 1 token on 1 session like local models. Local models rarely get to high utilization factor, they spend most of their time waiting.
If you had only batch inference and enough of it to fill the compute to 80% then you get cheaper local models.
Local models can absolutely run in batch, what are even talking about?
> If you had only batch inference and enough of it to fill the compute to 80% then you get cheaper local models.
Even if you ran sequentally, single session, a _finetuned_ tiny (8B) local model on narrow tasks would abolutely mog SOTAs, any of it - Fable, Opus, Sol you name it.
Can you share a bit more about your bifrost and llama swap setup? I’m facing memory constraints and am looking for a managed model solution that will help with hot swapping loaded models and stay-warm concurrency. Ideally with prioritization.
What do you want to know? Just start llama-swap with the models i have downloaded, add llama-swap as a provider in bifrost, expose the models you want and they become available in one single endpoint you can use in anything like opencode, openwebui or anything that speaks openai.
I have an M4 pro (48 GB ram) and I run Gemma 4 26b a4b at 52 tok/s and Qwen 3.5b a3b at 72 tok/s. Both 4bit quantized. These are enough for my needs and the performance is more than good enough. I'm not running the MLX version of the Gemma model, if I did the inference speed would likely be a bit better. I wouldn't use them for coding features though.
My perf sucks compared to yours. Added it to the post - same model averages 325 tok/s in processing prompts, and 34 tok/s in token generation. What am I doing wrong..?
Some examples (keep in mind this is all indefinitely free for me, no burning quota away):
1. Getting information (such as information about hardware unfamiliar to me) when not connected to the internet, which happens occasionally in my case.
2. Continuing to learn Rust by way of toy examples, puzzles, and comparing aspects of various solutions, for example from LeetCode.
3. Reformatting data, for example from a PDF to a markdown table, or converting receipt images to text.
4. Simple translation/explanation (e.g. I'm teaching my wife one of the languages I speak but sometimes may not know/have the words to explain the full nuance of a translated word).
5. Summarization. One of the webnovels I'm reading has some very boring parts I don't want to slog through, in those cases I simply make the LLM summarize that part and move on.
Etc., you get the idea. It's not unusable for coding, but it would make many mistakes when making a whole feature and the context lengths are limited to around 30k-40k tokens by my RAM. I could give it access to the web but I simply use an online model when I need that sort of thing, again partly due to the context limit.
Edit: The MLX version of Gemma 4 26b a4b does about 62 tok/s.
I'm the author - hello! Added to the post! Qwen averages 325 tok/s in processing prompts, and 34 tok/s in token generation. That isn't instant, but it's quick enough that I never really think about it.
Are you using the right configuration for your own CPU?
On a Laptop with 32 GB RAM and Iris Xe integrated graphic card, I get between 11-18 Tokens/Second with Qwen 3.8 27B and llama.cpp with sysl Intel optimisations. Same results with the vulkan back end, although sometimes it ends in weird segmentation faults due to the memory consumption.
I honestly wouldn’t bother with local models right now unless I either had a 5090 and was happy with running Qwen 3.8 27B, or a pair of DGX Sparks running DSv4 flash, or better, 2x6000 RTX Blackwells. Those are the kinds of rigs that the local model enthusiasts are running. With the GPU setups, you’re looking at generally >100tps generation in single stream, and >10k tps of prefill, so it’s snappier than Claude code, which somewhat makes up for it being dumber.
That said, it is really cool to be able to run an LLM on eg a Mac laptop. Just not a better experience on almost any metric for interactive use than eg Claude Code, beside privacy and guardrails.
>I honestly wouldn’t bother with local models right now unless I either had a 5090 and was happy with running Qwen 3.8 27B
How's the actual performance of Qwen 3.8 27B? On deepswe it supposedly performs slightly worse than gpt 5.6 luna high[1], but I can't help but think they've been benchmaxxed.
Not sure, I haven't run it, I've just been running DS V4 Flash non-stop since it came out, and that's replaced a lot of my Claude Code usage. People seem very impressed, though, it seems like it trades vram/world knowledge for extra thinking time, which I think is a good trade for local. tbf, I've heard luna's not great at coding. Fast and good for things like classifiers, summarization, though.
A friend and I were actually discussing today how benches show Luna Max at about par on coding with Sol Medium, but how it's nowhere near in reality. We were speculating that maybe it's because a lot of benches are best-of-n, and should probably be worst-of-n, because variance in performance is killer with large coding projects. Consistency is what lets you actually build on this stuff.
I took this thread and summarized it with Qwen3.6-35B-A3B, it had 1400 tps prefix and 60 tps completion. Very good performance. Using oMLX on MacBook M5 Pro 64GB.
It’s not. Do it as a hobby or for privacy but for performance just use a frontier model api. You’re paying less than cost for something that would take tens of thousands to set up locally.
That's not even remotely close to being true, even once you account for capex. You have to look at the actual usage, look at the token limits. Even if you're paying Anthropic $200k/month for scale-tier, you're going to blow through your token limits trying to run max output 24/7. Three users running Opus 4.8 at max non-stop will probably clean your monthly allowance from daddy Dario in less than a week.
With an 8x MI355x cluster at full tilt and including cooling, your power draw runs ~17kW. That's what it looks like when it's running full tilt. To be fair, hey that's pretty expensive. It does mean 8 multi-trillion parameter models unquantized running 24/7 without pause. And you get the full month like that, your monthly token limit is the time in a month. That cluster, the electrical upgrade, the cooling setup, and the electricity to run it all costs less in 2 months than your maximum affordance from Anthropic does in the same time period. Two billing cycles, and realistically it's more like two weeks. In 4 quarters you've wasted over a million. Like, what are we talking about here?
Now if you aren't using AI all that much, which is perfectly valid, and especially if you aren't using it at its absolute maximum, the story changes. Because even though at that point you're not paying nearly as much in electricity to run the cluster anymore, you still have the $300k+ capex to get the setup in the first place. But if we're not redlining it non-stop, then we're not really talking about performance anymore, are we? If your org never comes close to hitting token limits, it's probably because AI is rather marginal for you. Which again, is perfectly valid. I don't even use AI professionally.
Fact of the matter is, if your corp can justify the capex for a cluster and makes heavy use of AI, you are literally burning money by not having one in your building. The numbers are painfully obvious. Even deepseek isn't as cheap. This is before we get into things like LoRAs, custom inference pipelines, etc. which you know are kind of important if you actually care about model performance.
Here’s an experiment: purchase an anthropic pro max subscription for $200/m. Now go buy the hardware to run DeepSeek’s equivalent. In a year, who spent more?
In normal times in which hardware used to depreciate (lately that's not the case and HW even appreciates, but let's not get distracted), if you calculate only with depreciation costs, plus the fact that when you have such a setup, it'd take many 200$ subs to cover your lack of limits in the other, I think it'd not be a clear victory for any side.
If you just ask "who spent more in the first year" (100% depreciation) then even with 5-6 max accounts, buying HW will be a couple of times more expensive. But when does it make sense to ask that question?
Maybe the SotA models will need better hardware so your investment will not be useful after a year or you'd need very expensive upgrades? But then (as in Fable case) subscribers need to spend more too.
It’s not so clear after 5 years that you’ll come out ahead. You’ll have spent $20k. The apple computer owner will probably be running local models that are better than today’s frontier on the same hardware.
Idk where you live, but where I am running the M5 Ultra Mac Studio at max rated power 24/7 for a month costs C$42.
The considerations against Apple hardware are 1) hardware advancements 2) early access to the best models. But it’s really not that clear.
(The other guy who thought hosted models on openrouter are cheap has spent $100k in 5 years.)
> With an 8x MI355x cluster at full tilt and including cooling, your power draw runs ~17kW. That's what it looks like when it's running full tilt. To be fair, hey that's pretty expensive.
Pretty expensive is an understatement. You couldn’t buy one of these if you wanted to right now. If you could it would be multiple hundreds of thousands of dollars.
> It does mean 8 multi-trillion parameter models unquantized running 24/7 without pause
You can’t even run one unquantized multi-trillion parameter (>=2T) model on 8 x MI355x with enough context for concurrent users. I don’t know how you think it’s going to run 8 of them at the same time. Did you mean 8 concurrent sessions?
Your math is way off across this post. If replacing an Anthropic subscription for a whole company was as easy as buying a box for the office and then breaking even in 2 months, it wouldn’t be some little secret that we only discover in a comment online.
>You couldn’t buy one of these if you wanted to right now.
You can: https://www.exxactcorp.com/Exxact-TS4-149591758-E149591758 . You can get thousands of tps of GLM 5.3 output out of this thing, which grades around Opus 4.8. Payoff is around 1 year vs. spot prices on these GPUs, including power.
Ah thanks for the solid info, too bad. I'd seen them come up as a pretty good price for 6000 RTX's in the past, which seem generally pretty available, good source for those?
Yeah, they're good source. But the price for those GPUs is 5 figs even with the nvidia startup program nowadays. Also, I went back and looked. Most of my GPUs are actually from Central Computers who were great, but Exxact is real too. So "lots of" was inaccurate.
Also, the lead time I quoted was for individual 8x nodes.
I can't tell from the ad -- it says "supports" 8x MI350X GPUs, but does that mean "includes" 8x MI350X GPUs? For $300K I'd certainly hope so, but I'm assuming not.
A system with 4x RTX 6000s costs about $60K these days, and can (as you note) trade blows with Opus 4.8 if not Fable. In fact, it'll give you a better pelican than Fable 5.1, and in less time.
Ha fair, I'd definitely confirm with a salesperson before wiring them $300k. But most of the signs on the configurator seem to point to it including the GPUs? Not going to make 30k BTUs/hr of heat without the 8kw of GPUs.
Baseline yeah. But part of the reason you run open models is how much nicer fine tuning them is. Granted, you probably don't want to try and make LoRAs on a 4x RTX6000 setup, but you could if you really wanted to and there are other ways to modify models. And yes, if you're good at it, you can turn a piddly mid-range model that's only good at benchmarks into a heavyweight clanker (for a specific domain).
Well, they are if you're into animating pelicans. :-P But yes, in the general case Opus is a better match.
And Opus is no slouch. I'm satisfied that GLM 5.3 is just as strong as Opus. Z.AI has promised/bragged that they will be at Fable 5.0 level by the end of the year or early next year, and I don't see any reason to doubt them.
> Pretty expensive is an understatement. [...] If you could it would be multiple hundreds of thousands of dollars.
Obviously, I quantified both the operating expense and the capital expense in my post. What I find curious is that you're quoting me talking about the operating expenditure, and changing the topic to be about the buy-in like these are interchangeable things. You don't think that this is a crucial and important distinction?
> You couldn’t buy one of these if you wanted to right now.
You could have spent all of 5 seconds of searching rather than just assuming[1]. You're not buying an Nvidia Superpod™.
> You can’t even run one unquantized multi-trillion parameter (>=2T) model on 8 x MI355x with enough context for concurrent users.
That's certainly fair a point. Although in the English language, especially in legal contexts, the multi- prefix is used inclusively for fractional values. That is it's strictly >1, not >=2. IE an 18 month contract is a multi-year contract, or a $1.6 million dollar asset is a "multi-million" dollar asset. But this is uninteresting semantics.
You are right, but it also doesn't matter. The gap is just that big. You can run 1 single user of Kimi K3 and still not even come remotely close to the $70k or so that a single Opus 4.8 user can burn over the course of a month on left on max. An honestly lowballed amount I know from anecdote. The per-token cost is just really expensive.
> Your math is way off across this post.
You made one technical point above, one that doesn't ever arrive at a relevant rebuttal to the substance of my post. But please, I'd love to hear you elaborate, especially because I didn't actually give much math at all.
If you want math though, here's the math. Let's say you are paying a ridiculous amount of money for electricity, a price nobody in the US pays -- $2 per kilowatt hour. That's about 5x the average rate in California, 4x as in Hawai'i. 17kW @ $2/kWh * ~8766 hours in a year puts that cluster's electrical costs at just shy of ~$298k annually assuming it takes no breaks. Let's make matters worse and round that up to $300k. It's also assuming you didn't invest in a solar hookup for your building, which I don't know why you haven't at this point, especially if you're installing a CDU for your new cluster. 12 months of Claude burning $70k a month is $840k. For a buy in of, you know what, let's call it $500k. Why not? It still doesn't matter. The operating cost is so much lower it's paid for itself plus an additional $40k in the first year. Even at a ridiculous penalty in electricity that nobody pays, even overinflating the amount of money you'd pay for the cluster and the infrastructure to get it set up, it's not even remotely close for a single user where the gap is smaller (IE, you're not wasting "a million dollars" in a year by maxing out the $200k scaling limit every month)
You can of course trot out the point that oh, in 12 months this setup will be extremely outdated! It doesn't matter. If the work it was doing today was useful, it will be useful next year too. And with the rapidly encroaching diminishing returns from parameter scaling, you're probably going to be just fine for a while. Maybe grab a quantized version of a newer Chinese model at the end, before grabbing a newer generation of AMD node. Those MI400s are looking pretty sweet after all.
> If replacing an Anthropic subscription for a whole company was as easy as buying a box for the office and then breaking even in 2 months
If you're locked in, then you're locked in. But don't pretend like you're saving money. You're not.
> it wouldn’t be some little secret that we only discover in a comment online.
Why does this have you so nasty and defensive? It's not a "little secret" that running your own infrastructure is cheaper. Of course it is. You know what else is cheaper? Owning your own office building out in the sticks, rather than leasing part of one in the city. Not everybody can make that work, there are no free lunches after all.
History repeats, these same exact lines were rolled out ad nauseum during the cloud craze. Datacenters are businesses, not charities. Frontier companies rent quite a fair amount of their infrastructure. Even if they resold that compute below cost (they don't), there's a pretty steep cliff before the economics start to look attractive.
> You could have spent all of 5 seconds of searching rather than just assuming[1].
I guarantee this will not ship to you any time soon.
The current lead time on these GPUs in measured in years. If you didn't place an order for this a long time ago, it's not coming this year.
Being able to add it to an online configurator does not mean anything right now.
> 12 months of Claude burning $70k a month is $840k
Your math is completely useless with these arbitrary numbers pulled out of the air.
If you want to begin calculating payback period you'd need to look at token costs, cost per task, utilization rates, and so on.
> The operating cost is so much lower it's paid for itself plus an additional $40k in the first year.
You went from paying back in a couple months to paying back in a year but you still haven't even talked about tokens or concurrency.
You're also neglecting the fact that hosted tokens are going down in price at a rapid rate. If someone was paying $70K per month in tokens for Opus this month, that same level of compute is going to be much cheaper 12 months from now.
> Why does this have you so nasty and defensive?
Not nasty or defensive, just tired of these armchair claims that it's easy to go out and buy an 8 X MI355X box from people who obviously have no idea what the hardware lead time is like right now, or who haven't considered the actual math on token costs and payback times. You're still making a lot of claims without a single discussion of cost per task or token.
It does make me wonder how the hosted stuff is so cheap. For pretty much everything else, hosted/rented is more expensive but offers better convenience and flexibility. But for AI, even if you consider the total lifetime cost and are utilizing it heavily. You never break even by buying.
They're not cheap at all. I did one xhigh Qwen 3.8 27B agentic coding task last week via OpenRouter and it cost me like $10.
99% of the cost was in input tokens, I only used like 100k ish output tokens. It was a one shot task asking the agent to implement proxy injection to Guice. It did a pretty amazing job.
If you were to use hosted LLMs for a lot of agentic coding, a maxed out M5 Ultra Mac Studio would pay for itself in under a year.
I've been hosting Qwen3.8-27B myself. On my endpoint it's $0.30/1M in, $0.10 cache, $2.03 out - so those agent turns that re-send the same prefix get a lot cheaper when cache hits. UI at inference.tiyuvta.ai/app if you want to try it. Hosted is up to 210 tok/s and 280ms TTFT with reasoning off.
Qwen is weirdly expensive. Deepseek v4 flash is dirt cheap. You'd need at least 128gb of ram to run this model and in my experience, a days work with it costs around 80 cents.
So I ran the math, assuming the agent takes 75 turns per 200k context, with deepseek v4 flash it costs around $2.57 to reach 1M context in 375 turns. Cached input costs scale quadratically with # of agent turns.
Considering that I hit the 1M compaction multiple times per day with codex, it would definitely cost at least $5-8/day to use deepseek how I normally use codex.
Recent performance data on my M1 Max 32GB MacBook using oMLX. I have been working on identifying suitable model and config for my use case and system. Using a refactor and suggest improvements prompt for a specific Django code block using VSCode Cline extension.
Qwen3.8-27B-4bit generally runs out of output token before completing the task though excellent partial results.
Ornith-1.5-35B-A3B-MLX-4bit seems to get in the loop often specially with tool calls.
Qwen3.6-35B-A3B-mxfp4 seems to be optimal with speed and quality output.
I am going to test Qwen3.6-35B-A3B-4bit soon with same code block just to check my intuition that any derivatives don't seem to perform better than the originals.
I got 400 pp tps on a 10k token input. Your numbers seem suspiciously low, maybe the input was too short to measure properly? And this dense 27B is slow, the MoE A3B models get to 1000 tps.
Not enough for coding. 48GB is minimum for a non-lobotomized coding model like qwen, and you'd likely want 64GB to have long context and not kernel panic when Chrome opens.
You could run one of the smaller Gemma models to have a chatty Wikipedia.
It depends on your use case, but the smaller Gemma 4 models or qwen3.6:9b would probably run OK on that. I recommend trying it, even just for fun. It‘s easy with omlx.
> Running a large model locally comes down to one thing: how much RAM it actually needs in memory.
Not completely true. It's memory AND memory bandwidth. You can have 1tb of memory but if you have awful memory-bandwidth you'll also have slow tok/s. A3B helps with this, but so does MTP.
From my experience, you'd be better off running the dense 27b-mlx with MTP than the 3.6 version with A3B. You say your model is ~20GB of ram, but the 3.8:27b-mlx is 18GB and gets me very reasonable tok/s, and greater speed if you disable thinking when not required.
The dense 27b Qwen on M4 Pro has a prompt processing speed of around 125tok/s which makes it ok to ask a quick question but impossible to use in an agent, as processing the first prompt of the agent with the tools and instruction can easily be 10 000 tokens
In this case the 35b a3b makes sense as it has a PP speed of around 800tok/s
True. This then boils down to a quality vs speed decision. the 3.8 27b is far better than 3.6 A3B from my experience. I'm happy taking the speed hit, given local models aren't as intelligent as frontier models. Anything that can get me closer to my CC experience both in reasonable speed and intelligence is worth it. With that said CC can also be slow at times, so it's locally the difference in experience is not always noticeable.
> The main reason to run local: cloud APIs are rented land. They can change their pricing, hit your usage limits, or swap the model being served behind the scenes whenever they feel like it.
Quite a lot of "local doesn't work" in here - unfortunately, often with not much details about what the people actually want to use their models for. Which I'd be curious about.
I, personally, do use frontier models in the cloud for a lot of (meta-)cognitive analyses that are heavy enough to have me run against the limits of payed accounts regularly - so I'm neither a Luddite nor stingy with cash in this case.
However: I have pretty good experiences with local models as well. My solid but hardly extreme desktop (with one RX 9070 XT 16GB) mostly serves gemma4:12b and specialized models (embedding) to my local network. This is for general use like simple queries, simple code, reformatting and the like but also for two specific tasks that are permanently running:
a) It's connected to Home Assistant (as a second stage after very simple "turn light XY on" commands which get processed without LLM). So, I can mumble into my smartwatch "computer, how much gas do we have in the warp core and how much energy did the bussard collectors make from the cosmic dust today?" (or describe a more complex light scene or create an automation I want or whatever).
The phone transcribes that - with a local model on device - and fires it to the desktop who has agentic access to HA, looks through the sensors and data, sees that I've tagged my solar panels and battery with nerd vocabulary. It makes the right conclusion, converts a few units and gives me back a nice overview. All hands-free while I'm sitting on the toilet.
b) It's the LLM backend for a personal radio station run by a fleet of nerdy/quirky AI DJs who's archetypes are represented more than well enough in the latent space of the "small" model to produce funny results. The DJs can produce consistent, individual segments and programs, run a playlist that works well for me (based on multi-layered audio analysis that also uses local LLMs), respond to song wishes and generally produce much better recommendations than Spotify ever could for me. And you can also put multiple of them in the "studio" to create hilarious crossovers that you would not get from a commercial entity because the IP owners would rather shoot each other in the face.
All of this doesn't even max the available resources, so I can shovel F5-TTS into the VRAM as well and have all my DJs have good, locally created voices (or voice clones of Captain Picard and Han Solo, if I wanted to) based on zero-shot voice cloning.
--> Far from "unusable". It just depends on the task. And I neither have to hand my keys to the Navidrome server nor to my Smart Home to any entity outside my local network.
Apple is working from the 'desktop' up to beefy servers with 64GB+ RAM. Nvidia is working from the 'datacenter' down to beefy racks with terabytes of RAM.
There isn't really an overlap yet.
Individual Nvidia cards exist on desktops but they're not really oriented for regular inference so individual developers are left with Macs or datacenter resources as their options.
I experiment a lot with local LLMs, particularly small ones like Qwen3.5 4B and 9B. I have build multiple experiments to make harnesses that use these models for code generation, planning, local search, etc.
These are really good models but the harness has to be built around them. I have a ton of generated system prompts for specific purposes. Even parts of a SolidJS stack, for example Route management, has its own prompt. These are experiments but the results are real. If we build harnesses around small models, we can build a locally running WYSIWYG editor which works on plain text prompts.
The performance, in simple tokens/second, is not the most important factor. For many private data points, like emails, I would rather have a local graph based search and LLM on top where the harness is specific to problems like calendar, contacts, finance, etc.
I run all experiments on an 16GB M4 Mac Mini but coding agents building the harness are a mix of Codex, Claude Code and opencode.
Most people running local models would probably love to run larger models if only they had access to big enough hardware. I'm curious: to those of you running models locally, if there was a way to inference the model of your choice at a reasonable cost by effectively time-sharing a B300 rack through some privacy-protecting intermediary, would you consider that?
If there was a "Mullvad of GPU clouds", would that solve the privacy concerns?
yes, and it's already some offerings like that but they all cost a lot because they only good for "I have some idea of workload for N hours or days" lets rent it and run. That fine for some experimentation but if you think about renting something 24/7 even for example to share it with the friends that will cost at least 4x from any API prices as result (something like rtx 6000 48gb will cost ~$470/m).
runpod.io is essentially this. You can rent the hardware for cheap in small time slices. I do this whenever I need to do a lot of embeddings, fast. I have an agent skill that will estimate the optimum hardware to reserve for the time/price constraints of the job, and you can spin up temporary inference for cheap via their API as well.
> You do not know what these companies do with your data once they have it. They might limit how it gets used, they might sell it, they might expose it.
This is the burning question for me, what are they doing with our hard work.
I'd have thought that sherlocking a user's $10M business would be too high risk, given the billions at stake if real evidence of this happening was found.
However, OpenAI are currently being sued by Apple for trade secret theft, and the way it was done seems to be abundantly idiotic.
yes, share performance, numbers if you can, also i wonder if you figured out a way to do a 2way audio with local models, or even explored that. I have a very similar setup but not too happy with the token speed, will try omlx though !!!
Have a macmini m4 32G, not the pro version, previously everytime I tried local LLM is a bit disappointing, and I finally decide to not waste time and perhaps in the future invest a better hardware to server more modern and dense model
I am curious is what is the 80% request served by this setup, I was using it for OpenClaw which run serveral cron jobs that discover stuffs over the wide internet, check my support system's unanswered tickets, browser X and some social media for me to filter the valued ones(though I have to say even with GPT 5.6 sol, the quality is low for the timeline X sent to me)
Btw, Tailscale is quite cool and did a good job, I was using it to serve the local LLM and connct the openclaw on a Linux Machine to it.
You have tried Qwen 3.8 27B before coming to this conclusion, I hope? It's an incremental improvement over 3.6, but I mostly want to make sure you didn't just try running some old junker before coming to this conclusion.
I really like these show and tell style posts. I’m always curious how people have their setups and what tools they use. Also the blog has a nice theme and is easy to read.
I wanna get a desktop Mac for local ai so that I don’t turn my laptop into a delta 15k rpm fan when I run things.
For more complex or important tasks, costs, autonomy and privacy matter, but then so does performance/quality.
So I'm not completely convinced it's really worth it; but it's tempting!
reply