GLM 5.3 is probably the sweet spot open weights model if you want to go beyond deepseek flash or the new glm flash. I used it with pi and had a fairly good time, especially since it’s less touchy about cyber and whatnot than the US guys. It’s slightly behind Kimi in ability but it’s a lot easier to run it, I’d expect prices (and speed!) from third parties to be noticeably better.
Assuming you’re willing to drop a fat stack of cash on the upcoming Mac m5 ultra with 512 gb unified memory, you can even run it locally, quantized to 4 bit. Whether it’s even slightly reasonable, well, my wife would probably skin me alive but maybe yours is more understanding.
* Models are evolving quickly with high worldwide competition
* Hardware is evolving despite RAM shortages
Is investing a huge sum of money in equipment for local inference a wise use of money? Or are M5 Ultra and equivalently priced local inference hardware future-proof enough to be worth it relative to how the market is evolving? Maybe it’s all a question of what you’d spend otherwise on serverless or dedicated GPU spend…
It is absolutely not worth buying hardware to run models for purely (long term) cost reasons. For open weights models the economies of scale means the cloud beats local significantly and your payback time is like 10 years.
However there are other reasons (e.g. privacy) that might make it worth running locally for some people.
I think the privacy argument that keeps coming up is overrepresented. Certainly ZDR is enough for an absolute majority of use cases? I see so much talk about local inference but I doubt most of it has privacy as a valid argument (not arguing it doesn't exist). It's fun to do things locally though. I've tried it as well but cloud is just faster and cheaper.
These companies have displayed zero respect for everyone's intellectual property getting these models trained.
I think not giving them your complete trust is reasonable! I'm not saying zero trust, and ZDR is fine for most things but I understand the people who don't want to stream their whole codebase out token by token.
I have a Strix Halo and dual 32GB GPUs in my desktop, that sit idle right now, because the electricity to run them and to cool them in 110F weather Texas is currently experiencing pretty much nulls any savings I might see over getting better models from cloud providers. While I mostly use Claude or Codex with subscriptions for agentic work, for API use DeepSeek has usually been my go to, but now I guess it's GLM 5.3 or the Flash version. And, for security work that Anthropic or OpenAI models are likely to refuse, I've been using Kimi K3 (also via subscription, though their subscription is extremely stingy), but I guess GLM is now the one for that, too.
Anyway, yeah, even at the prices I spent on my local AI stuff (I bought before RAMpocalypse really kicked into gear, so I bought old server GPUs for about $350 each and the Strix Halo for a little over $2k) it was never going to pay for itself; I just like to tinker. But, I can't imagine spending today's prices for hardware for local AI.
When the memory shortage ends, I'll be down to the Apple Store (or, more likely, clicking refresh on the Apple outlet every few days). But, until then, there continues to be a glut of cheap and free models in the cloud that are better than anything I can run locally and they're faster, too.
If they "should" cost 4k in the sense of marginal cost, then you will be spending more running the same at home, because your home hardware will always be less efficient.
So far I don’t regret buying an M1 Max device with 32Gb of RAM. The models available for it keep getting better (running just about okay for interactive use) and 400 GB/s of bandwidth is still considered a lot.
The models are currently improving much faster than the hardware and this doesn’t seem to have plateaued yet.
That's basically the question I'm trying to answer.
If you're paying Anthropic or OpenAI to use their models, harness, governance, etc., I could see the local inference potentially coming out ahead. They're already starting to ratchet down what your money gets you on their platforms, and that can be expected to continue as the leaders of those companies continue to seek the road to the El Dorado that is being a trillionaire.*
If you're looking to get into the guts of AI development instead of having it handed to you by a provider, that's where it gets murky. I'm wanting to write some sort of agent that does things and get into making outputs consistent in the like, and I'm not sure whether to host something on GCP or buy an M5 Mac.
*Note: El Dorado is a mythical city and many people died trying to find it.
With competition we kind of have guarantee up to what providers can do, they don't have that much control, the most radical thing they can do is to go bankrupt.
Part of it is knowing that whatever sort of enshittification the cloud providers do, my local programming environment won’t ever be less effective than it is today locally. It’s the same reason my entire development stack from editor to compiler is open source. I don’t need to modify it today, but I always must retain the option to do so later.
There are several things I do in my life that only pay off in the event of a big disaster, like an extended internet outage, civil unrest, supply chain disruption, war, etc.
I like to be able to do the things I do even if offline for weeks.
I spent a lot of money for more flash in my iPad Pro so I can keep all of offline wikipedia and OSM in it, for example, along with tons of books. It’s sort of like being a digital prepper. (Being a prepper is a spectrum, from anyone who keeps food in their pantry to people building bunkers under their house - how much you invest is a personal prudence and threat modeling decision.)
Also, privacy. And when I got the Mac Studio the 512GB was only $15k, which is dirt cheap for that much VRAM.
> Part of it is knowing that whatever sort of enshittification the cloud providers do, my local programming environment won’t ever be less effective than it is today locally.
Is that true though? Many of the core LLMs need to be retrained as languages evolve to incorporate changes (language specifics, compilers, tooling, etc.). To some degree this can be handled via context injection in a variety do forms (agents looking up documentation and so on) but inevitably it’s not stationary in time, just as your OSS stack (probably) isn’t (depending on the languages, technologies, and use cases).
So your hardware is to some degree dependent on the good merit of groups like Z or Alibaba or whomever pushing out updated open weight models that dumped loads of capital into to train. You can keep using the existing models but at some point I suspect they’ll start to have more friction due to dated specs in language and so on. Again there are tuning and ways of layering this information on, and in theory you can even do some training on your own but I don’t think it’s as stationary as being portrayed here.
Those updated open weight models may not always be there (updated on new data). The usability of them is probably fairly long to be fair, but I suspect you’re going to see explosion in everything from libraries to languages etc due to LLMs so even the rate of change across your OSS stack may cause these models to be dated quite quickly, at least in the core model which will require layering fixes.
To be clear I’m on the fence thinking about much of the same issues and as close as I am to pulling the trigger, I keep thinking of very valid counter arguments as to why it’s me just wanting this thing I own. Which may be enough.
One could also run it locally on a used dual xeon (or amd-equivalent) server with 512GB RAM, albeit slower, if you have a useful workflow for it that's like "take this day's efforts and run it through various analysis agents", combined with giving it one-shot tasks/modules to build overnight. You would want a place like a garage or basement to put the server because it'll be loud.
Memory throughput performance is better than many equivalently priced (talking about the cheaper end of the used/refurb market here) single socket systems because each CPU is talking to its own banks of RAM.
Using an example of llama-server on a 256GB (each CPU is attached to 128GB) system running qwen 3.8-flash-next Q8, which takes up about 188GB with the model and context, it's not that the inference engine 'knows' about NUMA, but if you use --load-mode mlock to load the whole thing into memory at once at the time llama-server launches, the kernel will distribute it relatively evenly across the memory connected to both CPUs.
I have just built an Epyc with 512gb DDR4 3200 RAM for a "reasonable" price and I'm hoping to have a setup with GLM as the architect and Qwen 27b/Next Flash as the implementer. This is 1/5 of the price of the Mac, but also probably 1/5 of the speed lol.
I’ll be very curious what you get with DDR4. I also almost went that way. I have an Epyc DDR 5 rig and the best I see is 10 tok/s. Caveat being that’s at Q8 and a 4090 doing pre fill so it could be pushed up.
The surprising thing for me is how much work you will need to cool the banks if you’re near your memory ceiling. My memory starts soft throttling at about 74C (dies may be hotter, that’s the bank temp) and will turn down speed to try to stay below 80.
Happy to send my llama.cpp config settings if you want it.
I am getting 10t/s on unsloth's Q3kxl with 2x3090s@250w. It's enough for me for now. I will probably upgrade the GPUs down the line. DDR5 would have made the price of the machine double and I just wasn't prepared to pay that much.
I find 10 to be very usable. It’s not (that) interactive but it chews through tasks. I let Kimi churn away at 4 overnight and it gives good results that are ready for me in the morning.
Honestly I suspect neither of them will be performing terribly well but with DDR4 3200 RAM I wonder if you'll be counting tokens per second or seconds per token. I mean, you do at least get a lot of memory channels at least, compared to consumer PCs. I am curious to hear what performance you get, I feel there is not enough information out there on what different setups manage to eek out.
The fastest I was able to get my Threadripper 3960X + 2x 3090s + 256GB DDR4-3200 to run a 2-bit quant of GLM-5.2 was 8 TPS. I would expect to be in seconds-per-token territory for a pure-CPU 4-bit quant.
One thing I'd like to try is MoE offloading: I have 2x32 GiB of VRAM and 128 GiB of DDR5 running at 4800 MT/s (only 2 channels though). I've seen people post difficult to believe MoE offloading results albeit a decently long time ago with older models. Maybe there is a quant that would fit with MoE offloading?
That said, I am guessing my problem is not enough RAM - but this poor consumer platform struggles to do memory training with 128 GiB as it is.
Now I surely regret not having gotten Threadripper and 256 GiB of RAM in the before-times.
My measurement was with MoE offloading, but there's only so much you can keep on-GPU with a 200GB quant and 48GB of VRAM. It's hard to overcome the CPU/RAM bottleneck.
For what it's worth, all of my hardware was used; I think, all-in, I'm probably at around 3k-4k USD? Not cheap, but also not the worst for something relatively versatile.
With a 4-bit quant of GLM-5.2, I can get about 0.8-1.1 tok/s on an underclocked dual Xeon E5-2698 v4 with 512GiB of DDR4-2400. I think it was specifically a Q4_K_M quant. Of course, the time-to-first-token is absolutely atrocious.
Which is completely insane for a ten year old configuration.
What model are you interested in? DS Flash 0731@Q4KXL I'm about 25-30tps. Same as the new Qwen3.8 Flash Next. The new GLM 5.3Q3KXL at 10tps. I've got 2x3090s which I didn't mention in the original message.
Though I appreciate how good it is at "solid" grunt work and at that price (in fact I am paying the grandfathered subscription price; mostly).
I am planning to let go for my Claude AI subscription which I now use only for "planning" and maybe use that via Open Router as PAYG (at to try how it ends up). But god glm is bad at "talking" and "responding" anything prose. Not only quality but it's almost impossible to tune it and make it let go of its habits and biases and enthusiasms which often result in too many too and fro.
So I sometimes wonder at what point that starts becoming the cost and mental hassle. Maybe it's not there for me yet.
Well if you did get the m5 ultra could you obliterate the guardrails and then your wife can ask it pertinent but unsafe questions about how to punish you. Seems doable.
and even if you were making such a salary, the quesiton of if the investment on hardware to run llm's locally is still a big if, its OK if you buy the HW cause you'll use it and you get the extra capability as a nice extra, but doesnt make sense to spend so much when you could just get 200$ subs with almost infinite SOTA tokens a month etc (if you dont need the local/privacy aspects of it)
I'd like to ask Sam Altman if he still thinks that it's too dangerous to publish GPT-3. I mean, no one would use it, but what is his reasoning for not publishing it now, in 2026?
There are risks associated with releasing historical proprietary models that were not designed for open release:
- It is trivial to extract samples of the training data that was used, which can bolster existing lawsuits/foster new ones.
- Older models are not as safety-hardened, so it is easier to coax unsafe behaviour out of them, which is a PR risk.
- It may be possible to divulge proprietary secrets from the model (e.g. architectural details that may still be relevant).
For these reasons, and more, it's unlikely that GPT-3/similar models will be released until these concerns are no longer relevant (e.g. when they become a purely historic concern, similar to the open-sourcing of other proprietary software from decades ago).
I think it would be an important historical document as well. We are potentially looking at the dawn of AGI and one of the most important models ever created. Each model is also a kind of ultimate time capsule, containing a snapshot of the entire human collective mind. If you wanted to ask a 2002 person what they thought about future historical events you can just ask them directly.
> If you wanted to ask a 2002 person what they thought about future historical events you can just ask them directly.
The weights arent the truth tho, maybe a timecapsule-vhs but i wouldnt trust llm weights more than more hardcore deterministic media that might get preserved to infer facts from an era.
The companies doing the training are becoming the "winners" that are "rewriting history" as they train their models.
There’s not such a straightforward relationship between safety and model sis.
According to the book The Thinking Game, lower quality models at that time were considered less safe, because they could be easily tricked into doing harmful stuff. In the book, Dario (of Anthropic) was the head of safety at openAI and was responsible for pushing for 10x scaling in training to make the models safer.
It does make sense, a smart model is going to be way harder to jailbreak into telling me how to synthesize anthrax (or something).
Models are quite safe when they're useless, actually.
In the times of GPT-3 I'd scoff at the idea of an LLM doing any hacking; today, I'm running several AIs on my code before publishing, and they are finding (and demonstrating!) RCEs on my localhost server.
For example, one found a missing check in a third party JWT library which allowed full account takeover, which I'd have never even looked at.
Hence I don't believe a single word coming out of these people's mouths. Their "beliefs" are just marketing.
From today's perspective, it sure seems like it, probably because increased capabilities have generated a new kind of danger. Back then, they were worried about stuff like the model telling me dangerous knowledge.
I certainly think the labs have muddied the waters using safety for marketing, but that doesn't mean less capable models weren't more dangerous at one point.
Extremely weak justification on their part, bordering on trolling. It's just marketing.
Easy access to malicious information hasn't proven to be the disaster these anti-prophets proclaim. For the last ~3 decades of widespread internet and search engines, you could trivially find all sorts of information (drug synthesis, explosives, etc.), and we're just fine.
(Corollary: easy access to good information did not make non-intellectual non-curious people smarter. Easy access to information does not have the consequences people immediately assume.)
In some interviews, OAI mentioned that they didn't think that GPT-3.5 would be a success. They thought it would be a cool toy and they decided to launch it to see how users react. That means that they didn't think GPT-3.5 was intelligent enough. But somehow once GPT-3.5 became a huge hit, people conveniently ignored the anecdote, and started to believe that AGI had been eminent.
Maybe I'm reading too much between the lines, but I suspect the reason is to rub his nose in the duplicity or naivety depending on how generous you're feeling. Publishing the model would be a confession that he was wrong.
AI policy is being shaped somewhat by the things Sam and Dario say. So even if you're not feeling vindictive, it's probably good to keep a track record of the previous things they have said as a Bayesian prior. People who don't know better listen to these people, and maybe they shouldn't.
If you want to dunk on sam for "it's too dangerous to publish GPT-3", you hardly need the release of gpt-3 to prove your point. All the other open model releases already provide pretty good evidence. Moreover the fact that the model wasn't release hardly points to the fact that he wanted to save face or whatever. Most AI labs don't release their old proprietary models, so the fact that gpt-3 wasn't released tells us very little.
I previously posted that DS4Flash was _good_ but not _great_ on two DGX Sparks, but I have to say that GLM-5.3 is pretty amazing. It's been able to tackle all the random hard problems I've thrown at it and it has the intuition that DS4Flash seems to lack.
We're nowhere near a Fable-class model IMO, but things are going to get interesting in this next year.
It's early days, but GLM 5.3 Flash is the first local model that feels good enough to me to be a "main" model without debating whether each problem needs to be sent to a stronger model. DS4 Flash is good enough at implementing given a plan, but I wasn't always a fan of what it came up with when asked to plan something.
The good news is it can only get better from here.
What's very promising here is the number of tokens-vs-accuracy ratio. I am assuming their "output tokens" means tokens generated as part of thinking and any tool calls (what are referred to as "input tokens" from billing PoV by service providers). The Chinese models like Qwen3.8 and GLM 5.2 are insanely overthinking in our workloads (which are highly complex data analysis tasks). It's a factor of 3-4x over Opus and GPT models. Even with cheaper prices per 1M tokens, the cost ends up being higher, in some cases 2x. So this is very promising from GLM 5.3. Looking forward to trying it.
I'm starting to think Opus 4.8 is significantly smaller than most people assume.
If it's significantly larger than GLM 5.3 (I've heard some insane guesstimates out there like upwards of 5T params or more), that would prove rather embarrassing for Anthropic.
I hear the argument here, but isn't it possible it has dramatically more knowledge and when you get outside the common cases many of us use it for, it'll have completely different capabilities?
I feel like most benchmarks cluster on a reasonably limited area of human knowledge
It seems like there is tradeoff between model size and the need for tool use, which - in my mind - is quite costly in terms of time and tokens.
More detailed world knowledge requires an exponential increase in model size, but most knowledge can be acquired ad hoc using search or database queries. This will fail for questions where the model lacks the knowledge to ask the right questions, but maybe this could be solved by a handful small inquiry models with different knowledge encoded in their weights?
You can't compare models released 6+ months apart. GLM 5.2 was same architecture as 5.3 and not nearly as good. Takes time to build frontier intelligence and distill down to smaller sizes.
I really like how it doesn't have that Claude talk. It just does the thing without Claude's "load-bearing honesty." It's probably my favorite model to interact with, even if it isn't the best or most reliable.
My second favorite model by now is GLM 5.3 flash which is very capable of day to day task. I use it as the main model and GLM 5.3 for task that is more complex
Do you use it to write HTML/CSS? Javascript? C++? There's a huge difference in ways people use models and if you are not specific about it then your comment means nothing, unfortunately.
z.ai coder plan, both in opencode and direct API access. I use it for my side projects like https://github.com/scosman/Biscotti (on-device meeting transcription and summaries).
I have been only using GLM models since last December and have had the best experience without any drama about tokens and geopolitical restrictions. The quality has been great and I am doing more and more with the latest 5.3 and am really excited that consumer hardware will develop in the next few years where I can run these at home.
I find GLM 5.3 Flash more interesting than 5.3. The fact 5.3 does not have vision is kind of a deal breaker. Also 5.3 Flash seems to be better at making pretty UIs.
Have you guys been having a good experience with OpenRouter? I tried it out recently with Claude, and it cached no tokens, charging me $200 for one conversation of 11 messages.
I tried using deepseek v4 flash with OpenRouter. It switches between providers too eagerly which resets the cache. Then, each provider begins to rate limit me for providing so many uncached tokens, so it just keeps on switching providers. I'm paying for every token... why rate limit me? It was unusable compared to just using the official Deepseek provider which has a much better cache rate.
I have seen you advertise your website a few times. I like the idea of not having to trust the router, so I took some time out of my day to critique your website: https://files.catbox.moe/v68cf7.png
My visit to your website went like this:
1. Visit models page
2. Try to find GLM-5.3-Flash (which is among the ~5 models that 90% of people currently care about)
3. Give up scrolling (which would have taken OVER 50 SCROLLS!!!) and use Ctrl + F
4. Try to find input/output/cached price
5. Scroll all the way up to find out which column is what
6. Notice that output price is cut off
7. Notice that the scroll bar is over 100 scrolls further down the page
8. Use Shift + Wheel to scroll horizontally (most visitors probably won't know this trick)
9. Notice that cached price is missing
10. Conclude that this is probably not a serious offering and bounce
There are probably more issues later on, but this is how far I got.
I would suggest you to:
- Deslopify all pages that a user may visit before conversion
- List important models first (see OpenRouter rankings)
- Move the most important information (model name/input/output/cached price) to the left
- Disaggregate the prices per provider (maybe subtables per model? not sure)
- Measure cache hit rate and compute effective price per provider (see OpenRouter)
(- Optional: Fix the broken link on your HN profile page. Currently, the only way to get from this comment to your website is a search engine.)
How am I supposed to navigate around there? For example the pricing page is empty or is that how it was supposed to look? On models and providers pages there are lists but no way to filter or get any kind of meaningful info. Or is this a WIP/POC?
I doubt it - AWS hasn't added any non-western models since GLM 5 and MiniMax M2.5 in February, afaik. Might be a deal with OpenAI (GPT 5.4 was the first to be available via Bedrock, in April) or might just be that there isn't a lot of demand due to corporate skittishness around models trained in China.
how feasible is it build a SOTA specialized model for some use case e.g. deal sourcing by using this as pre-trained model or a LORA or similar pattern on top? Gonna shoot my shot at a billion dollar business
Not feasible. We’ve seen again and again that generalized models—somewhat surprisingly—dunk on specialized ones in almost all cases.
The first case of this I remember was Bloomberg thinking that their extensive private data about markets would give a home-trained model better performance in finance tasks. The next version of ChatGPT beat them.
With how often new versions of frontier models are released, you likely won’t finish your work before it’s obsolete. The labs have whole teams dedicated to specific getting training data for specific industries (finance is one), and more powerful generalized models make better decisions even without specialized knowledge.
Your best bet is to get really good at training for something and then sell your company to one of the frontier labs for their post-training efforts.
yes would also be interested in that - using knowledge distillation and other special knowledge sources to post-train on top of an open model like GLM-5.3. I was wondering that when Elon Musk tweeted "Specialist AI’s (single language, single area of knowledge) are another 100X" (src: https://x.com/elonmusk/status/2089968914596045178) - maybe he knows something regarding specialist model training the general public does not know?
Is it possible to fine tune this model and unlock / extend its cybersecurity capabilities? I'm scared that maybe we are not ready for an open-weight model with high cybersecurity skills.
You can fine tune a model from a year ago to get extended cyber capabilities. Fine-tunes dramatically increase capability in specific use cases and don't require a lot of investment. Attackers have been doing this for a while now, they aren't waiting for someone else to make them a security model.
I get how you feel, but it's too late to be concerned. The cat's out of the bag. It's like being scared of moving from the bronze age to the iron age... when everybody already knows how to make iron, and the raw materials are everywhere. People are already making iron spears. We need to make iron shields.
We need open-weight models that are good at finding security holes so we can apply them to all of our software by default, and close every possible security bug, before the attackers find them. Every piece of software in the world should be held for release until it's scanned by a high-powered security model.
This is the same debate we had in the 1990's when strong encryption was considered a munition and not allowed to be exported. This just made the world less secure. And it was pointless anyway, because you can't really stop it being developed and shared. Eventually good sense prevailed and now we all have strong encryption. The same thing applies to security bugs.
Stealth Ox-Alpha was GLM-5.3-Flash, and Ox-Alpha was a strong coder (way better than DS4Flash). With both GLM 5.3 and GLM 5.3 Flash, I don't think I'll need any better coding models. The main thing I need now is faster, not better.
Looking at HF, it looks like the unquantized version is half the size of glm-5.2 756 GB vs 1.51 TB. I wonder how they were able to optimize it this well
It's actually slightly more expensive ($0.50 vs $0.48), but there's a temporary 50% discount.
I've seen dozens of conversations about it in last 24 hours, and every major inference provided added in first 24 hours. I think it's gaining plenty of traction.
OpenCode Go is probably using quantized down DS4Flash. They outsourced to 3th party providers to keep the cost down, and being able to provide that $30 value (instead of the initial $60 > $15).
We saw the same issue with GLM 5.2 when they still published publicly who the providers are on their website. Most ran FP8 but one was doing FP4, so you had this issue where one moment you had the better FP8 and another session you had the FP4 provider.
You can check the internet archive, it was in the FAQ part before they hide/removed it. So if you looked up the providers, and the published quants, yea, ...
Given that a lot of complaints are coming from people that felt OpenCode Go Flash feel like a step down compared to old OpenCode Go/DeepSeek API directly, it smells of a quantized down provider is mixed in.
OpenCode Go is becoming less of a good deal by the month. I pretty much only use it for mimo 2.5 pro now, and everything else is either ollama or openrouter.
Not in my experience. Tasks that would normally cost $0.08 on DSV4-Flash have cost me $0.30+ on GLM-5.3-Flash. These costs are after Deepseek's recent increase. Also GLM-5.3-Flash is so slow compared to DSV4-Flash. I would be fine with GLM-5.3-Flash if it was cheaper and at the same speed as DSV4.
I use DSV4-Flash on Max through Deepseek's API. I have been using GLM-5.3-Flash on High through Openrouter which I thought had a 50% discount. I must be doing something wrong for the costs to be off this much.
I've been using it quite a bit too. My main complaint is that it can be really slow sometimes — like, really slow — and the speed feels pretty inconsistent.
I don't think that's right, or if it is, OpenRouter has incorrect data. Several Chinese companies (headquartered in China) have Singapore listed as their region on OR. And some companies, like Alibaba Cloud, have multiple regions listed.
I'm happy to be proven wrong, but this makes me think that the region is where the servers are, not where the HQ is.
I couldn't find any article that states z.ai has a data center in Singapore. There are stories of their new 1 GW data center in China though. Also, openrouter lists HQs on their providers page which matches the regions. https://openrouter.ai/providers
How much usage do you find you get on these kinda models (I know the pricing changes a bit) compared to a $20 sub say for Google AI Pro in anti gravity?
I hate how difficult it is to compare prices when looking at subscriptions.
Would $20 in open router, using models like GLM get me more or less?
I think it'd get you less than a $20 sub to any of the big three. I've used it on OpenRouter and found it kind of expensive for the results, but that might change now that it's open weight and other providers can host it/compete with Z.ai. For the work I did with it, I would've rather used DeepSeek V4 Flash just because it's more economical and still gives good results IMO.
Z.ai does have their own subscription, but I haven't used it because their privacy policy was pretty buns last time I checked.
Yeah I used Ox alpha earlier this week when it was free and deepseek flash when it was free on opencode. Both were great. Z.ai’s subscription doesn’t look very good versus the others.
I burn through my current Google AI pro sub for the week in about 2 and a half days so wanted something extra to add to it, but don’t want to buy any expensive ultra plan. Flash models have got me about 98% of what I need, but quotas are still a bit low.
I made a little project to calculate that. It gets the prices for API access and subscriptions, and calculates an average cost per token, for a given rate limit. So the cost is cheaper if you get a higher rate limit (cuz you get more tokens per dollar): https://codeberg.org/mutablecc/calculate-ai-cost
tl;dr API cost (openrouter) is always more expensive than a subscription (for the same given model). you should always use a subscription first and only go to API pricing if you run out of your subscription.
In terms of which subscription is best, different ones provide different models, different amounts of tokens, different rate limits. So it depends on what model you want and how much you need to use it. The frontier ones are always more expensive than open weight ones, but a few subscriptions are starting to include frontier models like GPT-5.6 Luna (which is a great deal but not necessarily the best price-per-performance).
I’ve been trying out GLM 5.3 Flash the past < 30 hours, and although I’m only running at Q3, it feels different than most 100B to 200B models I’ve run before... More measured, thorough in thinking, and has so far passed all my private tests (and at Q3).
just tested (zai-org/GLM-5.3-Flash via together.ai) against latest DeepSeek-V4-Flash for a very specific task and thought i'd report here...
- price: DS4 wins... $0.0235 vs $0.0242 for ten tasks
- latency: GLM wins... 108s total against 154s
this is for a personal use-case where i'm detecting ads in a written transcript. sticking with ds4-flash for now since latency is not a critical factor
Assuming you’re willing to drop a fat stack of cash on the upcoming Mac m5 ultra with 512 gb unified memory, you can even run it locally, quantized to 4 bit. Whether it’s even slightly reasonable, well, my wife would probably skin me alive but maybe yours is more understanding.
* LLM usage is new for the world
* Models are evolving quickly with high worldwide competition
* Hardware is evolving despite RAM shortages
Is investing a huge sum of money in equipment for local inference a wise use of money? Or are M5 Ultra and equivalently priced local inference hardware future-proof enough to be worth it relative to how the market is evolving? Maybe it’s all a question of what you’d spend otherwise on serverless or dedicated GPU spend…
However there are other reasons (e.g. privacy) that might make it worth running locally for some people.
I think not giving them your complete trust is reasonable! I'm not saying zero trust, and ZDR is fine for most things but I understand the people who don't want to stream their whole codebase out token by token.
Anyway, yeah, even at the prices I spent on my local AI stuff (I bought before RAMpocalypse really kicked into gear, so I bought old server GPUs for about $350 each and the Strix Halo for a little over $2k) it was never going to pay for itself; I just like to tinker. But, I can't imagine spending today's prices for hardware for local AI.
When the memory shortage ends, I'll be down to the Apple Store (or, more likely, clicking refresh on the Apple outlet every few days). But, until then, there continues to be a glut of cheap and free models in the cloud that are better than anything I can run locally and they're faster, too.
*$1000? $14,000? Who knows but everything in the middle there has been claimed.
The models are currently improving much faster than the hardware and this doesn’t seem to have plateaued yet.
It varies by model, but I’m getting 50-60 t/s with Qwen 3.6 35B and Qwen 3 coder 30B.
I’ve also used Qwen 3.8 27B but I get 10t/s on it.
It’s useable in some use cases, but I rely mostly on my $20 Claude subscription.
Could be a long time till gets released
If you're paying Anthropic or OpenAI to use their models, harness, governance, etc., I could see the local inference potentially coming out ahead. They're already starting to ratchet down what your money gets you on their platforms, and that can be expected to continue as the leaders of those companies continue to seek the road to the El Dorado that is being a trillionaire.*
If you're looking to get into the guts of AI development instead of having it handed to you by a provider, that's where it gets murky. I'm wanting to write some sort of agent that does things and get into making outputs consistent in the like, and I'm not sure whether to host something on GCP or buy an M5 Mac.
*Note: El Dorado is a mythical city and many people died trying to find it.
The object permanence of not having to reinvent the world every time a model gets sunsetted has value.
with open models, there is ecosystem/market of providers, where you can easily switch to provider you like
There are several things I do in my life that only pay off in the event of a big disaster, like an extended internet outage, civil unrest, supply chain disruption, war, etc.
I like to be able to do the things I do even if offline for weeks.
I spent a lot of money for more flash in my iPad Pro so I can keep all of offline wikipedia and OSM in it, for example, along with tons of books. It’s sort of like being a digital prepper. (Being a prepper is a spectrum, from anyone who keeps food in their pantry to people building bunkers under their house - how much you invest is a personal prudence and threat modeling decision.)
Also, privacy. And when I got the Mac Studio the 512GB was only $15k, which is dirt cheap for that much VRAM.
Is that true though? Many of the core LLMs need to be retrained as languages evolve to incorporate changes (language specifics, compilers, tooling, etc.). To some degree this can be handled via context injection in a variety do forms (agents looking up documentation and so on) but inevitably it’s not stationary in time, just as your OSS stack (probably) isn’t (depending on the languages, technologies, and use cases).
So your hardware is to some degree dependent on the good merit of groups like Z or Alibaba or whomever pushing out updated open weight models that dumped loads of capital into to train. You can keep using the existing models but at some point I suspect they’ll start to have more friction due to dated specs in language and so on. Again there are tuning and ways of layering this information on, and in theory you can even do some training on your own but I don’t think it’s as stationary as being portrayed here.
Those updated open weight models may not always be there (updated on new data). The usability of them is probably fairly long to be fair, but I suspect you’re going to see explosion in everything from libraries to languages etc due to LLMs so even the rate of change across your OSS stack may cause these models to be dated quite quickly, at least in the core model which will require layering fixes.
To be clear I’m on the fence thinking about much of the same issues and as close as I am to pulling the trigger, I keep thinking of very valid counter arguments as to why it’s me just wanting this thing I own. Which may be enough.
That said, when I bought my pair of Sparks, the best model I could run on it was GPT OSS 120B. That has an AA score of 24.
Today, the best model I can run on them is GLM 5.3 Flash at Q4, AA score 57. Just still out on GLM 5.3 mixed quant.
So from that perspective, they are many times better value than when I bought them, and will likely continue to increase in value.
Does inference make full use of the memory bandwidth in a NUMA system?
Using an example of llama-server on a 256GB (each CPU is attached to 128GB) system running qwen 3.8-flash-next Q8, which takes up about 188GB with the model and context, it's not that the inference engine 'knows' about NUMA, but if you use --load-mode mlock to load the whole thing into memory at once at the time llama-server launches, the kernel will distribute it relatively evenly across the memory connected to both CPUs.
The surprising thing for me is how much work you will need to cool the banks if you’re near your memory ceiling. My memory starts soft throttling at about 74C (dies may be hotter, that’s the bank temp) and will turn down speed to try to stay below 80.
Happy to send my llama.cpp config settings if you want it.
Temp wise, no throttling, surprisingly cool.
That said, I am guessing my problem is not enough RAM - but this poor consumer platform struggles to do memory training with 128 GiB as it is.
Now I surely regret not having gotten Threadripper and 256 GiB of RAM in the before-times.
For what it's worth, all of my hardware was used; I think, all-in, I'm probably at around 3k-4k USD? Not cheap, but also not the worst for something relatively versatile.
Which is completely insane for a ten year old configuration.
Though I appreciate how good it is at "solid" grunt work and at that price (in fact I am paying the grandfathered subscription price; mostly).
I am planning to let go for my Claude AI subscription which I now use only for "planning" and maybe use that via Open Router as PAYG (at to try how it ends up). But god glm is bad at "talking" and "responding" anything prose. Not only quality but it's almost impossible to tune it and make it let go of its habits and biases and enthusiasms which often result in too many too and fro.
So I sometimes wonder at what point that starts becoming the cost and mental hassle. Maybe it's not there for me yet.
- It is trivial to extract samples of the training data that was used, which can bolster existing lawsuits/foster new ones.
- Older models are not as safety-hardened, so it is easier to coax unsafe behaviour out of them, which is a PR risk.
- It may be possible to divulge proprietary secrets from the model (e.g. architectural details that may still be relevant).
For these reasons, and more, it's unlikely that GPT-3/similar models will be released until these concerns are no longer relevant (e.g. when they become a purely historic concern, similar to the open-sourcing of other proprietary software from decades ago).
The weights arent the truth tho, maybe a timecapsule-vhs but i wouldnt trust llm weights more than more hardcore deterministic media that might get preserved to infer facts from an era.
The companies doing the training are becoming the "winners" that are "rewriting history" as they train their models.
According to the book The Thinking Game, lower quality models at that time were considered less safe, because they could be easily tricked into doing harmful stuff. In the book, Dario (of Anthropic) was the head of safety at openAI and was responsible for pushing for 10x scaling in training to make the models safer.
It does make sense, a smart model is going to be way harder to jailbreak into telling me how to synthesize anthrax (or something).
In the times of GPT-3 I'd scoff at the idea of an LLM doing any hacking; today, I'm running several AIs on my code before publishing, and they are finding (and demonstrating!) RCEs on my localhost server.
For example, one found a missing check in a third party JWT library which allowed full account takeover, which I'd have never even looked at.
Hence I don't believe a single word coming out of these people's mouths. Their "beliefs" are just marketing.
I certainly think the labs have muddied the waters using safety for marketing, but that doesn't mean less capable models weren't more dangerous at one point.
Easy access to malicious information hasn't proven to be the disaster these anti-prophets proclaim. For the last ~3 decades of widespread internet and search engines, you could trivially find all sorts of information (drug synthesis, explosives, etc.), and we're just fine.
(Corollary: easy access to good information did not make non-intellectual non-curious people smarter. Easy access to information does not have the consequences people immediately assume.)
According to me, this is nonsense.
What's the point of publishing it when it'll likely be outclassed by gpt-oss?
AI policy is being shaped somewhat by the things Sam and Dario say. So even if you're not feeling vindictive, it's probably good to keep a track record of the previous things they have said as a Bayesian prior. People who don't know better listen to these people, and maybe they shouldn't.
I think the release of kimi k3 is definitely arguably dangerous, we're already seeing consequences of elite-tier cyberoffense capabilities.
I cannot stand using gpt-oss, but I miss some of the creative spark of GPT-3 davinci dearly.
We're nowhere near a Fable-class model IMO, but things are going to get interesting in this next year.
The good news is it can only get better from here.
If it's significantly larger than GLM 5.3 (I've heard some insane guesstimates out there like upwards of 5T params or more), that would prove rather embarrassing for Anthropic.
I feel like most benchmarks cluster on a reasonably limited area of human knowledge
You do pay for the tokens, but in theory on a smaller model each token is cheaper.
I should do that myself, come think of it.
(And 4.5 was smaller than 4)
Not really, in that you just work with different constraints.
Anthropic and US labs in general has maybe 100s to 1000s of GPUs per person to experiment. Zai and Chinese labs in general have 1-10.
The priorities are different.
Rather embarrassing indeed.
You implying its better than opus 5?
z-ai/glm-5.3: also Z.ai, Novita, Atlas Cloud, IO.NET
My visit to your website went like this:
1. Visit models page
2. Try to find GLM-5.3-Flash (which is among the ~5 models that 90% of people currently care about)
3. Give up scrolling (which would have taken OVER 50 SCROLLS!!!) and use Ctrl + F
4. Try to find input/output/cached price
5. Scroll all the way up to find out which column is what
6. Notice that output price is cut off
7. Notice that the scroll bar is over 100 scrolls further down the page
8. Use Shift + Wheel to scroll horizontally (most visitors probably won't know this trick)
9. Notice that cached price is missing
10. Conclude that this is probably not a serious offering and bounce
There are probably more issues later on, but this is how far I got.
I would suggest you to:
- Deslopify all pages that a user may visit before conversion
- List important models first (see OpenRouter rankings)
- Move the most important information (model name/input/output/cached price) to the left
- Disaggregate the prices per provider (maybe subtables per model? not sure)
- Measure cache hit rate and compute effective price per provider (see OpenRouter)
(- Optional: Fix the broken link on your HN profile page. Currently, the only way to get from this comment to your website is a search engine.)
https://trustedrouter.com/models
The first case of this I remember was Bloomberg thinking that their extensive private data about markets would give a home-trained model better performance in finance tasks. The next version of ChatGPT beat them.
With how often new versions of frontier models are released, you likely won’t finish your work before it’s obsolete. The labs have whole teams dedicated to specific getting training data for specific industries (finance is one), and more powerful generalized models make better decisions even without specialized knowledge.
Your best bet is to get really good at training for something and then sell your company to one of the frontier labs for their post-training efforts.
The insufferable gatekeeping of the US companies is actively contributing to computer insecurity at this point.
I get how you feel, but it's too late to be concerned. The cat's out of the bag. It's like being scared of moving from the bronze age to the iron age... when everybody already knows how to make iron, and the raw materials are everywhere. People are already making iron spears. We need to make iron shields.
We need open-weight models that are good at finding security holes so we can apply them to all of our software by default, and close every possible security bug, before the attackers find them. Every piece of software in the world should be held for release until it's scanned by a high-powered security model.
This is the same debate we had in the 1990's when strong encryption was considered a munition and not allowed to be exported. This just made the world less secure. And it was pointless anyway, because you can't really stop it being developed and shared. Eventually good sense prevailed and now we all have strong encryption. The same thing applies to security bugs.
I've seen dozens of conversations about it in last 24 hours, and every major inference provided added in first 24 hours. I think it's gaining plenty of traction.
We saw the same issue with GLM 5.2 when they still published publicly who the providers are on their website. Most ran FP8 but one was doing FP4, so you had this issue where one moment you had the better FP8 and another session you had the FP4 provider.
You can check the internet archive, it was in the FAQ part before they hide/removed it. So if you looked up the providers, and the published quants, yea, ...
Given that a lot of complaints are coming from people that felt OpenCode Go Flash feel like a step down compared to old OpenCode Go/DeepSeek API directly, it smells of a quantized down provider is mixed in.
I use DSV4-Flash on Max through Deepseek's API. I have been using GLM-5.3-Flash on High through Openrouter which I thought had a 50% discount. I must be doing something wrong for the costs to be off this much.
There are other providers with much faster inference, like BaseTen at >100t/s: https://openrouter.ai/z-ai/glm-5.3-flash#performance
I'm happy to be proven wrong, but this makes me think that the region is where the servers are, not where the HQ is.
Apparently, Cami Clark was tight with Eric Schmidt. Per unsealed documents, she seems to have pursued Epstein to invest in her "luxury porn" businesses, after a divorce & going bankrupt? Wild: https://www.wsj.com/tech/ai/claude-dario-amodei-wife-anthrop... / https://archive.vn/MJI7q
I hate how difficult it is to compare prices when looking at subscriptions.
Would $20 in open router, using models like GLM get me more or less?
Z.ai does have their own subscription, but I haven't used it because their privacy policy was pretty buns last time I checked.
I burn through my current Google AI pro sub for the week in about 2 and a half days so wanted something extra to add to it, but don’t want to buy any expensive ultra plan. Flash models have got me about 98% of what I need, but quotas are still a bit low.
What did you find objectionable? I looked at it when I subscribed almost a year ago and I was fine with it (e.g. they don't train on your API inputs).
tl;dr API cost (openrouter) is always more expensive than a subscription (for the same given model). you should always use a subscription first and only go to API pricing if you run out of your subscription.
In terms of which subscription is best, different ones provide different models, different amounts of tokens, different rate limits. So it depends on what model you want and how much you need to use it. The frontier ones are always more expensive than open weight ones, but a few subscriptions are starting to include frontier models like GPT-5.6 Luna (which is a great deal but not necessarily the best price-per-performance).
- price: DS4 wins... $0.0235 vs $0.0242 for ten tasks
- latency: GLM wins... 108s total against 154s
this is for a personal use-case where i'm detecting ads in a written transcript. sticking with ds4-flash for now since latency is not a critical factor