US chip export restrictions may actually be an advantage for China's AI Infrastructure. Chinese companies are forced to speed up developing their own AI chips
US chip export winners and losers:<p>Winners: Huawei, SMIC, CXMT,Chinese ASML-competitors, OpenAI, Anthropic, Amazon, Microsoft, Google, Meta.<p>Losers: Chinese AI labs, Nvidia, AMD, TSMC, Micron, SK Hynix, Samsung, Intel.<p>Any company that depends on Nvidia hardware such as OpenAI, Anthropic, AWS are winners. It means less competition for Nvidia chips and services. If you think Nvidia chips are expensive now, imagine if Chinese companies can buy them freely. Also for American AI labs, it also means they can stay ahead of Chinese AI labs in compute capacity.<p>The American hardware makers lost the lobby fight in Washington.
I wonder why Chinese AI labs are losers?<p>In the short term maybe yes, in the long term, maybe they are the winners, they can build on top of cheap inference stack and eventually win on pricing
It was evident that this will happen.<p>> Compared with our initial baseline on the same hardware, we achieved a 3× improvement in end-to-end serving performance, reaching hardware efficiency and per-token cost comparable to mainstream NVIDIA GPUs. This demonstrates that Chinese chips can support frontier-model inference efficiently and economically at scale.
you can also derive some stats from the ~10T tokens a day on 100k devices, 100M / device / day, but then one has to account for the multi-gpu model size, and I need coffee before I go there
Yup. That was really short sighted. And good for China. And actually the overall global market market since supply will augment and competition will decrease pricing as well.
Didn't everyone make fun of Jenson for saying exactly this?
Not sure, but there are definitely people around the president on both sides, some who think they can addict the Chinese to our silicon, like its the new opium war or something
China themselves recognize this. After Trump relaxed sanctions and allowed NVIDIA H200 sales to China on a case by case basis, the Chinese government stepped in to essentially block it!<p>In addition to Huawei who make the Ascend series that Ziphu are using, there are also at least a half dozen or so other Chinese companies also making their own AI accelerators.
It created demand that would not have been there without restrictions
We built a complete production-grade inference service from scratch on a cluster of more than 100,000 Chinese-made AI accelerators. All production inference for GLM-5.3-Flash runs on this system.
very few people comprehend - how much of an asteroid level event for western AI labs this is.<p>china has cheap abundant power, now they can make their own inference chips (which was supposed to be a chokepoint), their models yeah can be 6 months behind the frontier - but most people don't need frontier models - small models r more than enough.<p>my only wish was labs like Mistral would make their own inference chips or partner up eg with established / new chip makers or companies like Oxide.
Most of the people had kinda guessed this when they decided to provide 100 trillion tokens for free.
[dead]
Interesting that the tone of announcements between US and Chinese providers is converging.<p>GLM has in the past been more technical rather than speculation about future development on RSI etc.<p>Also curious whether those 100k accelerators are entirely locally made. If that's genuinely end to end on all components including lithography, memory, design etc then that is quite a feat.
Ziphu (who make GLM) use Huawei Ascend processors made by SMIC. Huawei use a combination of domestic memory from CXMT and leftover (pre-sanctions) memory from Samsung.<p>Just like the rest of the world, including the US (Intel, Micron), SMIC are currently using ASML lithography equipment (DUV, not EUV), but Shanghai Aishengna are now moving into early production with their own DUV machines, with SMIC and CXMT as early customers.<p>There is also a state sponsored Chinese EUV development underway.
Any details on the latest approach to distillation would also be very interesting.
I am surprised at the lack of open-weights models in the >35B, but <200B range. I keep thinking about devices like the NVIDIA Spark and AMD Ryzen Halo, which have their 128GB of combined memory, but there are so few models made for that range. Nearly all the open weights distillations are for larger customer bases with <24GB VRAM.
American exceptionalism states that America is special and unique so everyone else must be a copycat. American ai labs don't need this kind of optimization and fable will outright refuse to do it.
<p><pre><code> "We implemented a series of aggressive memory optimizations, including..."
</code></pre>
This whole thing sounds like industrial scale auto-research, but done by people who actually know what they are doing.
Plot twist: the GLM optimization agent figured out that it can hack and use NVIDIA GPUs on a US Cloud provider and make the inference 10x faster.
Time to tackle consumer GPUs next, since I’m not getting that Intel Arc B770.
I might be missing something but when I went to their site they are more expensive than Claude. Why would I pick GLM over Claude? Is it they just offer more tokens in their plans?
<p><pre><code> > Why would I pick GLM over Claude?
</code></pre>
To support the company that makes their model weights available for download, while Anthropic lobbies to restrict access.
What are you referring to? Given the audience, my instinct is to assume "plan" refers to the GLM Coding Plans, which are all cheaper than their Anthropic counterparts. As far as I can tell, the API costs are also all cheaper than their roughly equivalently capable Anthropic models.
Anthropic: 17 USD (pro), 100 USD (max)<p>GLM: 80 USD (pro), 168 USD (max) -> with "limited-time event" discount this becomes 56 USD and 117.6 USD<p>I also don't understand why are they so much costlier, and I would also like to give it a try.
The $17 figure is Anthropic's monthly cost if purchased annually. I'll use monthly numbers.<p>Anthropic's Pro is $20 and corresponds to Z.ai's Lite at $18<p>Anthropic's 5x Max is $100 and corresponds to Z.ai's Pro at $80<p>Anthropic's 20x Max is $200 and corresponds to Z.ai's Max at $168
Not to digress from the core argument of Claude vs GLM being open weights….<p>I have both plans. Claude monthly €20 and Z’s €18 monthly. Running GLM-5.3 high on their monthly plan will hit quotas absurdly fast compared to Opus 5 High on Claude code. It’s almost unusable for AI driven development. I ended up using the Z plan for using GLM-5.3 as a detailed security reviewer and adversarial feedback. For that, it is much better than Opus which will flag and bail out for even simple security tasks that are aimed at defense.
> Anthropic: 17 USD (pro), 100 USD (max) GLM: 80 USD (pro), 168 USD (max) -> with "limited-time event" discount this is 56 USD and 117.6 USD<p>GLM's "Max" plan is (was?) equivalent to 3x Claude's 20x ($200) plan.
because GLM does what Clauden't
For one you would have to use Claude if you pick it. But seriously there is no way for you to determine if one is a better offer than the other, when the usage/tokens/credits are vague, detached, and won't tell you much without trying both.
If only this infrastructure could handle all the traffic. I've tried using glm via z.ai - and it's a snail kind of slow.<p>And at the same time you have pretty strict limits to your usage, so in many cases you can't even let it work all night, as you will reach your limit faster than that.
That it's slow doesn't mean it can't handle the traffic, just that this speed is the optimal tradeoff to them. They benefit from serving more tokens by exploiting parallelism across users at a lower number of tokens per second per user, instead of serving each individual user as quickly as possible. When there's a drop in traffic, they probably shut down GPUs rather than giving you higher speed.
Given the huge amount of money being spent on AI chips in the US, what prevents US AI labs from doing the same level of software optimization? It could be a solve for some of the capacity constraints.
They have already been doing it for months <a href="https://openai.com/index/openai-broadcom-jalapeno-inference-chip/" rel="nofollow">https://openai.com/index/openai-broadcom-jalapeno-inference-...</a> . OpenAI on their custom chip brought up lightspeed deepseek as experiment by using AI in the exact same way as this zAI blogpost. And the kernel optimization contests/etc have all been havily done through AI based optimization loops for half a year+.
> what prevents US AI labs from doing the same level of software optimization?<p>Because they don’t have to. Most of the time money would buy you newest and/or more hardwares so there’s low/minimal interest to optimize the code or approach.
> As we develop GLM, the model sometimes exhibits capabilities that surprise us<p>Creators of known unreliable programs be surprised their programs are unreliable.
I'm not feeling any of this speed optimization; it's dog slow.<p>Signed, a customer.
This is a really funny sounding post. They sound like they just found out that increasing your automation gives you increased capabilities at faster speeds. They also sound like they just realized AI makes hard things easier.<p>But what really kills me is the idea that these companies are using Python for production inference. I mean really? Have you seen how bloated and slow Python is? Do global locks really sound like a strategy for fast dynamic computation?
Most of the fastest inference and training code in production today is written in Python. There are no global locks on the GPU except the ones you put there
It's not that they "just found out" - what they are saying is that while they were previously dogfooding because it's good practice, now that their models are so much stronger they are using them because it helps accelerate.<p>If you look at how many years the whole NVIDIA and CUDA ecosystem has been evolving, it's certainly impressive how they've just stood up and optimized this CUDA-free 100,000 node cluster in just a few months.
Python acts as an orchestrator of accelerator libraries and does none of the inference math directly
Someone tell this man about vLLM!
> Have you seen how bloated and slow Python is?<p>Yes, but it's calling C code.
Necessity is the mother of invention. The shortsighted protections put on chips, etc., by the US has forced Chinese AI industry to adapt or die. Guess what their response to this fitness function has been? Kudos to Z.ai on their inventions and excellent write-up, which reads like humans wrote it.
Wouldn't it be refreshing if OpenAI and Anthropic were this open, and spelled out how they were using their own models during development and rollout?!<p>All I can recall reading from OpenAI about what they have actually done in the name of "RSI" is using one of their models to help automate the training process.
Different angle on the same model: the full GLM-5.3 (744B MoE, 4-bit experts, 434 GB on disk) runs on a single MacBook Pro M5 Max with 128 GB
by streaming the experts from NVMe SSDs instead of keeping them in memory.<p>One drive gives about 2 tok/s; striped across four drives it reaches 3.5 tok/s with byte-identical output, and our best internal build with a not-yet-published patch does 4.2.<p>Method and numbers: <a href="https://github.com/argonautlabsai/argodrive" rel="nofollow">https://github.com/argonautlabsai/argodrive</a> (built on antirez/ds4).
[dead]
[dead]
[flagged]
Given the rate of improvement, why is this deranged?
because the rate of improvement is fairly stalled?
Do you have anything that proves this one way or another that isn't based on vibes or shoddy benchmarks?
You're tragically misinformed; it isn't. Several metrics are actually growing exponentially. But if you want emprical information, you can just have al look at the nature of the late AI incidents.<p>Ironically, many benchmarks being maxxed out, and quite quickly, so new ones have to be created.
The AI "incidents" are pure marketing ploys to get free word of mouth. Like what you're doing.
If you knew what "exponentially" means, you probably wouldn't be saying that.
<p><pre><code> Several metrics are actually growing exponentially.
</code></pre>
Power consumption and water consuption are the obvious ones. What are the others?
[dead]
[dead]
Kind of but there's a lot of improvements available to the competitors catching up
Yeah, it's been over a week since a Millennium Problem was solved. AI has hit a wall.
It was not solved. ~OpenAI~ Buckmaster and Alpöge found one (or a few) singularities in the forced version of the Navier-Stokes equations. Then magically 2 weeks later OpenAI found them too. Again, I am not saying this is not a great feat. I am just saying that everyone should be a bit more careful when making statements about RSI.
if you can be replaced by an algorithm, how useful were you really?
[dead]
[dead]
> This is known as Recursive Self-Improvement, or RSI.<p>Some call this "The singularity" (e.g. Hinton).<p>This is actually a core danger postulated by the, let's call it, "worrying" scenario - see AI 2027 (to be clear, I think its timeline is not realistic).<p>> Statements dreamed up by the utterly deranged.<p>Evidently, and tragically, it will take catastrophes to show that deranged are the ones deriding the worried crowd.
[flagged]
I was gonna ask how people found their coding plans, and realized, have they massively ramped up the prices? Seems the middle plan is ~$80/month now, didn't that used to be like $20/month? Cheapest plan is ~$20/month currently.<p>They must have hit really hard scaling limits if the prices were hiked so much so quickly.
I paid $360 annual for Max plan and currently averaging about 1BN tokens a day with their frontier GLM-5.3 model. This was clearly unsustainable for them and they've dropped this package.
1 billion tokens a day?!! I've done a lot of work these past 2 weeks with GLM-5.3. Like, a lot. And I've just passed 300 million tokens in total.<p>Can I ask where are you using all those tokens?
Well, there's essentially two major ways to use these models: Pair programming or fully autonomous fire-and-forget code generation. The second strategy needs essentially zero input, so the number of tokens you can blow is practically only limited by API speed.
Something like this I guess: <a href="https://youtu.be/U-Rqv9dOB1U" rel="nofollow">https://youtu.be/U-Rqv9dOB1U</a>
I have 3-5 agent harnesses with large context windows working on different applications concurrently.
Share the resulting code from any one of those please? I've tried so many times to find a setup that facilitates parallel work + high quality results, but it's just impossible regardless of harness or model. Leave the agents alone for too long, and the entire thing just balloons out of control, and next you know you're sitting there with half a million LOC where 80% isn't even needed.
Most of them are not public, but a fun thing I did was a mario cli game - <a href="https://github.com/Daviey/mario/" rel="nofollow">https://github.com/Daviey/mario/</a> (or `ssh mario.baby`).<p>I now exclusively use <a href="https://omp.sh/" rel="nofollow">https://omp.sh/</a> as my harness:<p>I set it up so it never works in the main branch so subagents etc don't step on each others toes, and only merges back when complete:
<a href="https://github.com/Daviey/mario/blob/main/.omp/hooks/pre/worktree-guard.ts" rel="nofollow">https://github.com/Daviey/mario/blob/main/.omp/hooks/pre/wor...</a><p>A good AGENTS.md is essential:
<a href="https://github.com/Daviey/mario/blob/main/AGENTS.md" rel="nofollow">https://github.com/Daviey/mario/blob/main/AGENTS.md</a><p>I then provide specifications for what I want, making sure it is unit tested.
That's easy to do with many agents independently told to find bugs in a large codebase.
300M for two weeks is surprisingly low. What are you doing that need so few tokens?
I also have a legacy pro plan and the only limitation is if you are trying to work in the morning from Europe because you are in the 3x usage overlapping China time but after 12 or so you basically can run it at least for me at least 3 parallel sessions all the time.
Their plans are still worth it if you use their models. You can see how many tokens you can except to get based on plan here: <a href="https://docs.z.ai/devpack/overview#estimated-token-allowance" rel="nofollow">https://docs.z.ai/devpack/overview#estimated-token-allowance</a><p>The max plan will provide ~1,100 USD of GLM-5.3 or ~260 USD of GLM-5.3-flash per month for 168 USD. I can personally attest to these numbers through omp (~97% cache hit rate).<p>Unless you are able to highly parallelize (your work, you won't be able to hit your hourly or weekly quota using the flash model simply because it's so slow.<p>They give you ~3x more flash tokens, which maybe comes out to ~2x more actual work after accounting for the extra thinking it does to achieve the same result. The mental model, for not getting angry, is 5.3 is fast mode by default, and you can disable fast mode for 2x the work output at 1/3-1/10th the speed.<p>They're serving me 5.3 at ~40 tok/s and 5.3-flash at 30 tok/s (according to omp).
That table assumes cache hit rate of 95% or better. Am I understanding this correctly that people really are doing such repetitive prompts (compared to each other, across the concurrent user base at that time) that only 5% or less need actually be computed by the intended LLM?<p>That is shocking. Is it per-token I wonder?
Every tool call is essentially entire prompt so far sent again with the response and that's why cache rates are so high for agentic workloads.
This really bites when using expensive models since most models are 1/10 for cached input.
If you are using their coding plan for coding, then yes you can easily hit such cache rates, with a good harness.<p>I’m getting 97%.
The way I look at it, their coding plan doesn’t retain data or use it for training making it one of the cheaper plans for me.<p><a href="https://docs.z.ai/legal-agreement/privacy-policy" rel="nofollow">https://docs.z.ai/legal-agreement/privacy-policy</a>
>I was gonna ask how people found their coding plans<p>Very good - but I'm on a legacy plan. And coming up on a renewal that would put me on the watered down current plan. But with 50% legacy discount think it may be worthwhile. If I go to a competitor I'd be paying market rate.<p>>They must have hit really hard scaling limits if the prices were hiked so much so quickly.<p>Not really scaling - their plans were initially comically subsidized even more so than what the western providers are doing. More advert for an upstart than commercially priced.
Way to restrictive in terms of tokens provided. I am on their largest plan, and quickly run into their limits. And that is using it selectively in addition to codex.
Yeah it went from a great deal to unviable compared to other providers imo. They really need to find a healthy middle ground
It just gives a taste of what we are all going to have to pay soon, once the model providers actually have to make money. And the era of "let's charge a dollar for every 10 dollars running the infra actually costs" is rapidly coming to an end.<p>And you can bet GLM is still ridiculously subsidized, just not as ridiculously as Anthropic and OpenAI.
What provider are you using currently?
It's hard to know, since no one advertises the actual token limits (partially cause they're prolly complex / adaptive). So it seems much more likely that they just offer different pricing tiers than you're used to. Like, the $80 plan is still ~$80 of subscription quota, regardless of what else is offered.<p>For [API usage](<a href="https://openrouter.ai/z-ai/glm-5.3-flash#providers" rel="nofollow">https://openrouter.ai/z-ai/glm-5.3-flash#providers</a>) they charge a bit more than the very cheapest providers of GLM-5.3-Flash, but not so much that a big price difference would make sense.
This article left me with one immediate question: "WTF is GLM?".<p>Honestly, I have no idea what z.ai is either (I'm aware of an AI-enabled editor called Zed, but that's under zed.dev), so it's a bit presumptuous from them to assume that everyone is familiar with their product...
z.ai is a fairly well known AI lab out of China and their GLM models are probably the most popular outside of Anthropic or OpenAI’s. I don’t think it’s presumptuous for them to not introduce themselves in a post on their own blog, I think you’re just a bit out of the loop here.
I don't get the outrage. Do you post this kind of stuff on every topic on hackernews that you are not knowledgeable about?
Maybe my post sounded harsher than I intended, and yeah, it's probably on me that I'm not familiar with GLM. Actually the other major Chinese LLM Kimi does ring a bell, maybe it's because three-letter acronyms are a dime a dozen and annoy me because I'm confronted with them regularly at work too (people at my company seem to <i>love</i> acronyms), but that's obviously on me too...
> Maybe my post sounded harsher than I intended<p>Appreciate the clarification. For me it was the "F" in "WTF" that tipped me. Other than that, it's more than fair for you to not know what GLM is. Things are moving so fast that I would be surprised if anyone can keep track of it all. Cheers, have a grand day!
It didn't read as harsh. Only unaware and you broadcasted that you don't have the decency to do basic searches.
It's presumptuous for them to assume that a reader of their blog is familiar with their product?<p>Also I feel like the obvious way to read the very first sentence is that GLM is a language model<p>> As we develop GLM, the model sometimes exhibits capabilities that surprise us
It's only the top open-weights LLM in the world,<p><a href="https://artificialanalysis.ai/#intelligence-category-tabs" rel="nofollow">https://artificialanalysis.ai/#intelligence-category-tabs</a>
Ziphu, aka Z.ai, is the company that makes GLM (a very competitive Chinese LLM).<p>Why would you be reading their corporate blog posts if you don't even know who they are?!
A ai model family similar to Codex, Gemini or Claude.<p>Where GLM-5.3-Flash is the newest "small / fast" model.
>As we develop GLM, the model sometimes exhibits capabilities that surprise us, and even unsettle us.<p>Come on now<p>Also, why would they introduce themselves on their own blog?
Well, other than the infrastructure they got from illegally routing millions of paying customers' requests through Anthropic's Opus 4.8 in a distillation attack...
Anthropic infringed the copyright of basically every author on the planet: <a href="https://www.anthropiccopyrightsettlement.com/" rel="nofollow">https://www.anthropiccopyrightsettlement.com/</a><p>No real reason to respect any terms they might want to impose. Besides, if you want to break TOS, just have an agent do it; "everyone" running these things agrees there's no corporate or moral liability for what your AI does.
That is such a canard, IMO. FWIW, Anthropic and OpenAI encrypt "thinking" token outputs in their models, while Chinese labs don't. If anything, it's more likely that everyone is using open-weight models in their synthetic training data generation pipelines. It's way easier to distill from logits than it is to distill from hard tokens.<p><a href="https://x.com/EricSimons/status/2099252922098061714" rel="nofollow">https://x.com/EricSimons/status/2099252922098061714</a>
We weep for Dario, that he had to suffer such a devastating attack against his Terms of Service.
What is "illegal" about it?
breaking Anthropic TOS and misleading users
Are you joking...? Sorry if so! Just in case: It's illegal in both the PRC and the USA.<p>In the PRC, they[1] leaked tons of national secrets on the PRC's latest AI campaigns, the inner workings of their "opinion monitoring" (read: performative panopticon) and "stability" (read: violent oppression) departments, Chengdu's whole CCTV network, direct-energy weapons plans, espionage activities in Syria to hunt down Uyghur refugees, and god knows what else that Anthropic didn't divulge to us common folk.<p>In the US, it's very clearly an attempt to rip off a competitor. I'm not sure how else you could possibly see it. Even if you're a distillation fan in general (which A. why and B. plz don't), they did this through a network of Japanese and Signaporean shell accounts, presumably at least some of which were abusing Anthropic's subscription service in a ToS double-whammy, as it would be exorbitantly expensive otherwise. They also had to hack around Anthropic's API to get CoT traces, which seems impossible to explain away as anything innocent.<p>I've been beating the "China isn't necessarily an enemy, it's gonna take us all to handle AI" drum for literally years, but this attack was just... gross. Gross in scale and gross in arrogance. Not a good sign for the dawning alignment crisis, to say the least :(<p>TL;DR: Use these services if you want, but know that you're supporting aggressive escalations and companies that very clearly don't give a flying fuck about violating the law, much less your ToS. So... buyer beware, I guess.<p>[1]: For clarity, Z.ai was not alone in this, nor were they most egregious attack -- Moonshot.ai (kimi) took that coveted prize. DeepSeek was involved, too.
What does any of this have to do with the legality of distilling Claude?<p>> use these services if you want, but know that you're supporting aggressive escalations and companies that very clearly don't give a flying fuck about violating the law, much less your ToS<p>From my European point of view the same risk/concerns apply when using US providers
I'm always wondering when "distillation" comes up how feasible it is, or if it's just BS.<p>The Antrophic article mentions "16 million" conversations, GLM models are in the 700-300 billion parameter ranges and while the frontier sizes aren't know but Gemini suggests Astra and Mythos are at around 10 trillion. That'd amount to extracting 40k parameters per conversation without a lot of errors if it was just a distillation (from an unknown source/algorithm as opposed to distilling your own model).<p>Now, I can imagine these conversations being used as a verification step that they're not missing stuff in their training, and that their models are capable of most of the same things, but that's mostly confirming that they've stolen the same data from the public as Antrophic/OpenAI has stolen already.<p>Or am I missing something here that makes real "distillation" feasible?
> alignment crisis<p>Alignment is meaningless; as you've noticed, humans aren't all that "morally aligned".<p>If the tool needs safety measures it should be kept in a safe enclosure like we do with CNC machines, furnaces, and so on.
You didn't explain why it's illegal or why distillation is bad.
Nulla poena sine lege?
Source for 1? Are we sure those aren't hallucinations?
Like Anthropic and OpenAI are? After all, didn't they distill all the information in the world into their model(s)?<p>I mean, if they get to distill other's IP, why can't others distill their IP?
Yes, wont somebody please think of the shareholders whose IP had been stolen...
[dead]
I have very little sympathy for thieves who get robbed of the goods they have stolen.
If you understand what they have achieved here, then the notion that they are bottle-necked on training data is absurd.<p>I wonder how you imagine that China built their own space station? Reliant on using American made duct tape, perhaps?<p>Do you realize how reasoning models are being trained nowadays? You design/build simulation environments to run agents in, with the environment providing the RLVR "verification" scoring. So why won't Ziphu use GLM to build their own RL training environments? Do you think they are not doing this?
Eh, even if this was true, then they're merely stealing from thieves. Anthropic did break a ToS or two to get training data themselves.
[dead]