upvote
No one runs SOTA models 24/7 for individual use or even for a single household or small business, whereas you can run your own hardware basically 24/7 for AI inference.

With the new DeepSeek V4 series and its uniquely memory-light KV cache you can even extend this to parallel inference in order to hide memory bandwidth bottlenecks and increase compute intensity.

This is perhaps not so useful on a 128GB or 96GB RAM Apple Silicon device (I've seen recent reports of DS4 runs with even one agent flow hitting serious thermal and power limits on these devices, so increasing compute intensity will probably not be helpful there) but it will become useful with 64GB devices or lower that have to stream from a slow disk, or with things like the DGX Spark or to a lesser extent Strix Halo, that greatly overprovision compute while being bottlenecked on memory bandwidth.

reply
deepseek v4 flash on mlx at 1m context runs at 20 t/s decode on a mac studio m3 ultra with 512gb of RAM
reply
What is everyone running DeepSeek v4 Flash with?!

It’s currently unsupported on Llama.cpp and vllm doesn’t support GPU+CPU MoE, so unless all of you have an array of DGX Sparks in your bedroom, what’s the secret sauce?!

reply
https://www.github.com/antirez/ds4 (from Antirez of Redis fame) runs a 2-bit quant on Apple Silicon hardware and 96GB or 128GB RAM.
reply
Just because you read it on a github repo doesn't make it true, it also doesn't take into account cpu temps and inevitable throttling you'll encounter.
reply
i ran it on my own device haha

i don't comprehend why people are in such disbelief at how much better this stuff runs on a mac studio than on NVIDIA hardware with 1/5th the VRAM. look, what can i say? NVIDIA is a bigger rip off than Apple is!

reply
Which is good, because Nvidia pulling a Micron and ceasing consumer hardware production is right around the corner.
reply
deleted
reply
API prices are most likely not subsidised. A brief look at openrouter can tell you that. There are plenty of providers that have 0 reason to subsidise that sell models at roughly the same average price. So the model works for them (or they wouldn't do it otherwise).
reply
They are subsidized, heavily. This is simple math, there are lots of reasons to subsidize. Please go look up the hardware requirements to run your favorite model and a given tok/ps then multiple that by 86400 (seconds in a day) then divide that by 1mm and multiple by the $ per mm tokens, then ask yourself if there's any possibility they could be profitable or even close to break even.

You are going off vibes alone, this is easily verified, please go verify.

What makes you think they have zero reason to subsidize, because the providers aren't a household names you assume they wouldn't operate at a loss? Whats your logic here? You make no sense.

reply
Serving a single user is likely not profitable, but total throughput rises a lot when serving many concurrent users, because the same weights can be used to generate tokens for all users at once, which increases efficiency.

Also, a lot of money is being made on input tokens and cached tokens, which are much cheaper to compute.

DeepSeek published their math for serving the V3/R1 models. They were 535% profitable: https://github.com/deepseek-ai/open-infra-index/blob/main/20...

reply
The amounts of API tokens many large companies are using through, say AWS bedrock are quite high. We've seen leaks on the bills for real world use cases. It's not unreasonable to see normal individual subscriptions as possibly subsidized.... but do we think someone like Anthropic is going to be subsidizing 7, 8, or even 9 figures monthly bills from megacorps? Because said megacorps will swap out to a competitor immediately, so your subsidy is unlikely to lead to loyalty or anything.

If Anthropic and OpenAI are subsidizing the metered API usage, their model is going to end up just as successful as MoviePass. They are burning enough money on the training costs already.

reply
Large companies are paying an arm and a leg, but I'm still certain even at $15.00 per million tokens they are not profitible.

If you have a machine running at 150 tok/ps you can only make $5820 a month at $15 per 1mm running 24/7. It costs a hell of a lot more than 6k a month to run Claude 4.7 @ 150 tok/ps on that machine 24/7.

This math is a bit off, because you have input tokens too, but regardless its still not profitable especially for how long it takes to turn around a request and the caching is probably not all that profitable.

reply
You are all over this thread, but you have no idea how inference works, and it's obvious. Your napkin math is off because you don't know what to add up, you lack the necessary background. And yet you persist and reply all over this thread. I don't get it.

Serving models on dedicated hardware is not the same as your at home 150t/s thing. Inference is measured in thousands of tokens / s in aggregate (i.e. for all the sessions in parallel). That's how they make money.

reply
Anthropic and OpenAI make money on API calls, margins have been reported in public filings. Subs are subsidized.
reply
That's not possible, read my comment above. These are private companies, there are no public filings regarding their profitability in any sense. You're just making things up.

If you have a machine running at 150 tok/ps you can only make $5820 a month at $15 per 1mm running 24/7. It costs a hell of a lot more than 6k a month to run Claude 4.7 @ 150 tok/ps on that machine 24/7.

This math is a bit off, because you have input tokens too, but regardless its still not profitable especially for how long it takes to turn around a request and the caching is probably not all that profitable.

reply
You're forgetting a critical factor: concurrency. If a given hardware serves a single request at 150 tokens/s, it can also serve 20-30 requests at 100 tokens/s. Suddenly your $5K becomes $100K/month, enough to recoup the cost of the hardware in a year or so.

The reason it works: each time you read the model (memory bound) to calculate the next token, you can also update multiple requests (compute bound) while at it. It's also much more energy-efficient per token.

[1] https://aimultiple.com/gpu-benchmark

reply
Interesting I didn't know about this, but it makes sense after reading the article. They are benchmarking on a single GPU on a 20bb param model. Does it scale across 60 H100s over NVLink/NVSwitch. I would be interested to see those benchmarks.

The idea that everyone is spinning up a $2 million in GPUs to scan their email inbox, search the web or avoid learning something is still ridiculous to me regardless.

reply
Training to be artisanal coder now.
reply
It cost 20-30k a month to run Kimi 2.6. The tokens are sold for $3 per mm.

Not if you're OK with 4-bit quantization. More like $30K-$50K one time.

Spring for 8 RTX6000s instead of 4, and you can use the full-precision K2.6 weights ( https://github.com/local-inference-lab/rtx6kpro/blob/master/... ).

reply
RTX 6000 Pro retails for $10k so an 8x is $80k before anything else in the computer, and long-context will have... pretty bad performance (20+ seconds of waiting before any tokens come out), but it's true it technically works.

I don't think cloud models are going away; the hardware for good perf is expensive and higher param count models will remain smarter for a looong time. Even if the hardware cost for kind-of-usable perf fell to only $10k, cloud ones will be way faster and you'd need a lot of tokens to break even.

reply
> I don't think cloud models are going away; the hardware for good perf is expensive

I think local AI will win in its niche by repurposing users' existing hardware, especially as cloud hardware itself gets increasingly bottlenecked in all sorts of ways and the price of cloud tokens rises. You don't have to care about "bad" performance when you've got dedicated hardware that runs your workloads 24/7. Time-critical work that also requires the latest and greatest model can stay on the cloud, but a vast amount of AI work just isn't that critical.

reply
Users do not have an existing $80k of hardware, are not going to buy $80k of hardware for worse performance than paying $100/month, and models are continuing to grow in size while memory grows in price.
reply
You said you need $80k in hardware for "good performance". I'm saying the local AI inference workflow will be a lot more flexible about performance than that, and can get away with something vastly cheaper and in line with what the user owns already.
reply
> paying $100/month

There will not ever be a monthly subscription for LLM tokens. The economics isn't there.

Local tokens will always be cheaper.

reply
What's the basis for saying local tokens will always be cheaper? As others have outlined, LLMs serving one user at a time are pretty expensive, but concurrent users become much more cost-effective (assuming there's enough RAM for the contexts). If "local" to you means ~10 hours daily use by a team of employees, the company still has to balance against cloud services that can amortize non-recurring costs over 24 hours of service per day.
reply
Why wouldn't a team of employees be able to run AI workloads 24/7? Not all workloads are time sensitive.
reply
"I think"

Well your thinking is completely vibes based and not cemented in any reality I exist in.

reply
Other sites beckon.
reply
> higher param count models will remain smarter for a looong time

They're not smarter, they just know more stuff.

You probably don't need knowledge about Pokemon or the Diamond Sutra in your enterprise coding LLM.

The "smarts" comes from post-training, especially around tool use.

reply
If the smarts came from post-training, we could show significant gains by doing that post-training again for previous generations of models. But we know that isn’t happening - effective post training is necessary but not sufficient for model performance.
reply
If 8 x RTX 6000 is getting you 20s before initial token, how are cloud vendors doing this?
reply
4-bit quantization is native for Kimi 2.x series.
reply
You're right, I was thinking of Qwen. K2.6 will run at UD-Q2_K_XL precision on 4x RTX6000 boards, but I have no idea if it's worthwhile.
reply
Posts like this are so funny to me. I'm staring at a mountain of old hardware right now that cost about $20k ten years ago. I have to pay someone now to come haul it away. What makes you think the current new hardware won't end up with the same fate.

> Just write your own fkin code people

Bro is nostalgic for googling random stack overflow threads for 10 days to figure out a bug the agent fixes in an hour.

reply
Do you have any old laptop ram?
reply
It's old rack mounts. Only one of them has some ECC DDR4 worth something.
reply
I'm just saying that agent that can fix your bugs actually cost $100-150 an hour to run and you're getting it essentially for $200.00 a month.

The cost of cloud compute actually hasn't gone down for old hardware all that much, it still costs $500.00 a year rent 4 core i7700k that's 10 years old. Don't expect much more valuable hardware, like modern GPUs to deflate in price all that quickly.

There's 3 fabs in the world that make ddr7 and they aren't going to be selling their stock to consumers going forward, it will be purchased by datacenters almost entirely and stay in them until EOL.

Your brain is going to atrophy (this is proven), they'll raise the price to something thats closer to break even and you'll be forced to pay it because you no longer have those muscles.

reply
The architectural problems I deal with day in day out leave no room for atrophy. This is just cope.
reply
You're going to see major cope once that bargain $200/month plan goes away, and every person or company that has embedded these services into their workflows gets to see their actual costs.
reply
Have you actually tried this stuff or are you just saying stuff you hear on the internet?
reply
> two 4090s is not consumer grade

I think that is a very narrow perspective. Enormous numbers of consumers own $50,000 cars, but a pair of $2000 GPUs is "not consumer"?

I agree with your view that cheap tokens on SOTA are a trap-- people should use local AI or no AI.

reply
> Enormous numbers of consumers own $50,000 cars, but a pair of $2000 GPUs is "not consumer"?

$50k is a median priced car in the US. I'd guess >99.9% of people do not own $4000 of GPUs. I consider myself a computer person and I dont think I even own $4000 of computer hardware in total

reply
> I consider myself a computer person and I dont think I even own $4000 of computer hardware in total

A top-spec MacBook Pro is >$4k, so I assure you that plenty of computer people do own $4k of computer hardware.

Hell, most tech folks are wandering around with a ~$1k smartphone in their pocket too.

reply
Fwiw you can finance a car over something like 7 years now. So a lot of people will be paying like $750 per month, not $50k lump sum.
reply
Plenty of gamers own serious GPU rigs that are reusable (at least to some extent) for local AI inference. That's almost certainly more than 0.1% of the populatiom.
reply
I guess I wasn't clear-- I wasn't so much making the point people do own $4000 in GPUs (though I suspect you are massively underestimating the number who do, also before the current market conditions this would have been more like $2500 in gpus...), but they certainly could per the evidence of car ownership.

A car is super useful, so is an AI. But even if we decide cars are incomparably more useful a great many people pay much more than $4000 over the minimum viable car, and that's money that could be deployed to secure access to private, secure, and autonomous AI facilities. A few thousand dollars in computing is consumer hardware, or at least could easily be with more reason and awareness driving adoption.

People spend a LOT of money in things less useful than local copy of qwen3.6-27b can be.

reply
I would still question what usefulness there is with a local model even with 10k in GPUs. I certainly haven't seen any great uses myself from these smaller models (<500 parameters) except claims from people who are totally enamored with AI and basically anything output from an LLM impresses them like a toddler who's entertained by the sound their velcro shoes makes.
reply
Probably you're focused on coding agents? I bet someone could use that kind of hardware to filter snarky comments
reply
Here is an example-- I'm running hermes + qwen3.6-27b on a workstation GPU (an older RTX A6000 which gets 55tok/s, though people run this model on more limited hardware).

A friend an I had previously worked on an entropy extraction scheme and he recently got around to making a writeup about our work: https://wuille.net/posts/binomial-randomness-extractors/

I instructed the agent to read the URL, implement the technique in C++ for 32-bit registers, then make a SIMD version that interleaves several extractors in parallel for better performance. It implemented it (not hard since there was an implementation there that it read), then wrote more extensive tests. Then it vectorized it. It got confused a few times during debugging because the algorithm uses some number theory tricks so that overflows of intermediate products don't matter and it was obviously trained a lot on ordinary code were such overflows are usually fatal. I instructed it to comment the code explaining why the overflows are fine and had it continue which mostly solved its confusion.

It successfully got the initial 12MB/s scalar implementation to about 48MB/s. Then I told it to keep optimizing until it reaches 100MB/s. I came back the next day and it had stopped after 6 hours when it achieved just over 100MB/s. Reading what it did: it went off looking at disassembly, figured out what hardware it was running on, and reading microarch timing tables online and made some better decisions, tried a lot of things that didn't work, etc. (And of course, the implementation is correct).

I'm pretty skeptical about AI and borderline hateful of many people who (ab)use it and are deluded by it-- but I think this experience shows that a small local model can be objectively useful.

(oh and this experience was also while I only had the model running at 19tok/s)

Running the model in a loop where it can get feedback from actually testing stuff allows you to make progress in spite of making many mistakes.

I could have done this work myself but I didn't have to and I certainly spent less time checking in and prodding it than it would have taken me to do it. In my case I wondered how much faster parallel extractors using SIMD might be-- an idle curiosity that would have gone unanswered if not for the AI.

reply
This is maybe the first time Ive seen someone claim to do something useful with such a small model.

Congrats, but you're in the 0.0001% thats not just frying their brains, fapping to their local models or doing various magic tricks like a toddler entertained by playing with velcro.

At the end of the day you lost an opportunity to improve yourself and excercise your brain, maybe the opportunity cost is worth it idk, but Im going to keep taking things slow.

Handmade swiss watches > mass manufactured immitations. Handmade clothes > walmart clothes.

reply
Sounds like you're coping for the vendor lock-in you cornered yourself into.
reply
This is a change that's been happening gradually over time-- I don't think I could have done this on a local model that could run on a consumer class gpu a couple months ago.

There are plenty of other uses that people have been making for a long time-- e.g. I know someone who uses a fine tuned local model to sort their incoming email and scan their outgoing messages for accidental privacy leaks.

I don't agree with your assessment on an opportunity lost-- I got my reps in on the original work, the AI gave an incremental step forward which made the whole exercise somewhat more valuable to me with minimal additional cost. I think this improves the cost vs benefit in a way that makes me more likely to try other pointless activities, knowing that when I run out of gas I can toss it to AI to try some variations.

Sometimes you're also 27 steps deep on a nested subproblem and you're really just trying to solve sometime. Even in finr craftsmanship not every step needs to be about maximum craftsmanship. :) Sometimes it's just good to get something done.

I think this is much like any other tool. One can carve furniture using only hand tools, but the benefits of a router are hard to dispute. Both approaches exist in the world and sometimes both are used in concert.

As far as people frying their brains with AI -- you don't need local models for that, plenty of people are driving themselves into deep personally and socially destructive delusion just using the chat interfaces.

reply
I do think post training smaller open source models for very narrow tasks is largely overlooked and there'll be lots of value there if one puts in the effort. However, in a lot of cases we're just compeleting a circle back to deterministic behavior at 1000x the memory/compute requirements just to avoid writing regex.

I agree with you, there's a way to use them responsibly like your router anology, I just think most aren't doing this correctly and its a slippery slope. I'll contend that you probably have used them responsibly in your example.

reply
[flagged]
reply