Even if it didn’t happen here, it was still the case that it was going to happen going forward. It was always going to end like this. Invest in the hardware companies, not the model companies.
But that said: K3 is not a distilled version of Fable or Sol. Fable has been barely available and Sol was just released! Moreover, K3 is superior to both models in some domains, according to user scoring on the Arena.
API distillation can’t give you these results anyway. All it is useful for is bootstrapping RL in new domains to get past the “cold start” problem faster. By far, what matters more is the quality and variety of RL environments the model learns from.
This behavior is exactly what you'd expect from a model distilled from Claude.
There's a detailed analysis of K3's ambiguous identity here: https://github.com/rgreenblatt/which_claude_is_k3/blob/main/...
This analysis observed K3 identifies itself as Claude approximately 15% of the time.
K3 reproduces Claude's correct current model id, which the real Claude models themselves do not emit. This suggests K3 was trained on Claude data labeled with deployment metadata (API logs, tagged synthetic data), rather than Claude's chat outputs.
And there's an entire Reddit thread discussing Kimi's similarities with Claude https://www.reddit.com/r/LocalLLaMA/comments/1m2w5ge/did_kim...
This analysis shows K3 and Opus/Fable have unexpected correlated outputs https://typebulb.com/u/lab/you-re-relatively-right/full
Either way, there's probably no significant portion of Mythos/Fable or Sol in there as OP has stated.
Here's another report of K3 identifying itself as Claude https://x.com/Sauers_/status/2077842686459981901
And an analysis showing the self-identity distribution for K3 and other models https://x.com/RyanGreenblatt/status/2078663148509544589
"I'm actually Claude - not Kimi". https://x.com/PimDeWitte/status/2077884701470040083
I regret to inform you that it is, in fact, real and from their own website - you don’t even need to try hard to reproduce it. https://x.com/PimDeWitte/status/2078105292965912690
lmao this is so funny, if you ask Kimi K3 for something with an empty system prompt it will consistently think of itself as Claude https://x.com/__alula/status/2078359305741275445
"I genuinely believe I'm Claude based on everything in my training" https://x.com/williawa/status/2077869021589033002
another "I'm actually Claude - not Kimi", including the system prompt https://x.com/jchudnov/status/2078661564803207406/photo/1
this is not hard to repro, just use a system prompt that doesn't mention the model name.
that said, if they bootstrapped with opus 4.6 convo sft data they had sitting around... so what?
Instead of spending 12-18 months building their own robust harnesses and painstakingly creating quality training data (which is what Anthropic and OpenAI did), they distilled Anthropic's models to complete the hardest part for them. Chinese labs cut 18 months of work into 6 months, and are now head-to-head with their American counterparts.
Anthropic tried to complain about this unauthorized "token theft", but they burned too much public trust with BS restrictions and users don't care. The US government is too busy fighting a war to help. Chinese labs are offering cheap, free & open-weight models; exactly what users want, and users will overlook any questionable methods Chinese labs used to achieve this.
The cope is incredible. There's people in this thread in denial that Moonshot AI is trained on exfiltrated Anthropic's model output, even when shown substantial evidence this has been happening since Kimi 2.X
Chinese labs were even paying an absurd $0.01 per Opus tool call trace, to get the quantity of training data needed.
Kimi K3 has reached the point of RSI, and no longer need teaching data from Anthropic/OpenAI models to improve. K3 can generate and iterate on its own training data, and self-improve.
We witnessed the most extensive industrial espionage campaign, probably ever, and nobody cares at all that it happened.
more like silk than capacitors
if, again, your model is that RSI will be beneficial, why wouldn't making it available to all unlock more benefit globally than not doing that
People are going to be gobsmacked when, in our lifetime, China becomes a world power comparable to the U.S. Probably still poorer per capita, but at Spain/Italy levels, not third world country levels. And they’ll be shocked at the implications of that on the world economy, migration patterns, etc. There will be fields where China is a global leader, and Americans and Europeans will have to learn Chinese and move there, or else be stuck in some satellite office of a Chinese company. We’re all in Europe circa 1895 not realizing the behemoth America will become in WWI.
Some day China can pioneer in science or technology but the current claim about Chinese companies leading AI development is ridiculous given the evidence of distillation and the fact that like 95 percent of science that lead to the current state of AI happened in either North America or Europe.
To be honest if you want to list academic papers that lead to the current AI models the majority is either done by Google Research or sponsored by Google.
Thus far the US has not really chosen to go the Chinese rare-earth method yet. The problem with distillation attacks is the end result is everyone who is not doing them is going to deal with some kind of regulation whether it's complete loss of access, or the amount of control you'll have to give up to access them will be ridiculous.
Sort of like the "stealing music is fine" but "lets freak out now that it's producing visual art", in the end the entire thing is a social construct. Whether this is treated as theft or "business as usual" is entirely societal.
Eventually the gap will close, unless there's a major breakthrough that hasn't been made yet.
Taxes on AI subscriptions or AI capable hardware, to financially compensate IP holders for (potential) IP theft, could very well arrive in the near future, once the industry is mature.
If this shocks you and sounds preposterous, I'll remind you that in several EU countries, we still pay extra taxes on any and all storage mediums and on devices with built-in storage (tapes, CDs, DVDs, HDDs, SSDs, tablets, phones, etc) simply because they can be used to store pirated content, decisions based on laws from 50-100 years ago, and the money goes to the national unions and associations of music and arts IP holders. It's basically a lobby pushed and government legalized extortion racket that no voter agrees with or can change but has no choice but to conform either way.
So I guarantee you in the future, it will be the same for AI subscriptions and hardware capable of running LLMs locally. Every time you purchase a Claude or ChatGPT subscription, an Nvidia GPU, Intel/AMD SoC PC or an Apple/Qualcomm powered smartphone, you'll pay a government enforced tax to the likes of Sony, Axel Springer, etc. for licensing their IP, whether you want to or not. In the EU at least. US maybe not.
Under the settlement, Anthropic was forced to delete the pirated data they were training on.
Chinese labs can still train on pirated data. I doubt the Chinese models operate under similar licensing agreements.
The payment was for illegally downloading copyrighted material, not training. Training was explicitly ruled to be fair use.
Training on legally acquired / licensed data is potentially fair use.
And other district courts don't agree on this. The US district court for Delaware recently rejected a fair use defense for the use of copyrighted works to train AI. https://www.reedsmith.com/articles/court-ai-fair-use-thomson...
There are more cases in the pipeline. The massive NYT vs OpenAI is still ongoing. Nothing will be "settled" until this makes its way to the Supreme Court or Congress steps in.
> I doubt the Chinese models operate under similar licensing agreements.
US corps likely pay licenses when afraid to be sued, or have troubles getting that data, otherwise they just take data, which was demonstrated many times. The same apply to Chinese corps, alibaba totally can be sued in US.
I believe mechanics is following: US corp sues Chinese, asks for preliminary injunction to stop selling product for example if there is strong evidence some IP for example was stolen etc. Then they litigate, and settle somehow.
> What are the most high-profile examples of the "tons" of lawsuits resulting in Chinese companies being banned from doing business in the U.S.? Isn’t it usually more action by the government - executive orders, etc?
In response to "What are the most high-profile examples of lawsuits resulting in Chinese companies being banned from doing business in the U.S.", the one example given was from 2 years ago of a ban that lasted for 2 weeks (separate from its 2019 onward government bans)?
However, if the claim is that companies (including Chinese) can face significant fines from IP lawsuits, I agree.
Chinese labs can freely train on pirated material, which is a structural advantage.
Meanwhile, Chinese labs are speeding in a different county. Everyone knows they are speeding, yet the sheriff won't pull them over, so they just keep doing it.
This lax enforcement gives Chinese labs a structural advantage over American ones.
Do you purport to know for a fact that they're no longer training on the data they'd pirated? Because I highly doubt that.
Destruction of Materials: In addition to the monetary compensation, Anthropic has agreed to destroy the two libraries that allegedly contain the pirated works, as well as any derivative copies originating from those sources. Anthropic must certify in writing to class counsel that the destruction has been completed and that the allegedly infringing materials are permanently removed from its systems.
The libraries in question were Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi).
If Anthropic is somehow training models on deleted data, I'd be quite impressed.
there is much less intellectual property in China so it’s not ‘theft’ (as you can’t put property on information)
I'm not aiming for a what about kickflip here: I'm saying we need to either agree on some rules or stop crying foul. Maybe the coherent legal theory is that neural networks and intellectual property don't interact. That would be weird but it would be consistent, a market could price it, I could do coding stuff and know if I was illegaling.
But this weird gerrymander that no judge will really rule on in an emphatic way is like, bad for the planet, bad for markets, bad business.
There are a lot of reasons to look forward to DeepSeek Huggingface drop kicking the unambiguous frontier weights in like, November, but I think my favorite one will be "who's distilling now bitch?"
But the American AI companies only let you query their models if you first sign a contract to not train on the output.
It's hypocrisy and unfair, but I think there's a strong legal argument for it.
Of course China can simply decline to assist in enforcing that contract... But I would expect US courts to do their best to.
Now THAT'S doing some heavy lifting lmao. The vast, vast, VAST majority of the original datasets were from pirated books and the like. Also, arguably a robots.txt is the exact mechanism to follow to do the mass GET-ing, yet the AI cos choose time and time and time again to simply ignore it and be as abusive as they possibly fucking can
And there's been significant legal consequences as a result
> Also, arguably a robots.txt is the exact mechanism to follow to do the mass GET-ing
You're free to argue this of course, but the courts have largely rejected it already pre LLMs. See for example hiQ Labs v. LinkedIn
Or do you just mean that US courts don't have enough teeth to prevent Chinese companies from violating contracts? On that I agree.
The US is already publicizing the way they are using Claude with Palantir for war gaming purposes. It’s a matter of national defense. Contract law has no meaning here.
That right there is the problem.
Maybe today. I doubt it tomorrow. Legal and not legal, largely, has to answer to the population sooner or later. Ultimately, humanity decides legality. And I don't think the frontier labs will get a pass from humanity in the midterm, let alone the long term. I think you'll see the rules change towards something more "intent" driven. And there's absolutely no difference in intent between Frontier labs and everyone chasing them.
Frontier labs just want the door closed behind them, as do their investors, because they know the money will never be recouped if others can do the same magic tricks.
Situation right now seems more like a fragile detente: if you got a Hill staffer drunk and hounded him long enough he'd probably be like "God damnit the market will fucking tank if we don't get these two IPOs out north of a trillion. And don't even get me started on how I'm going to sell Chinese AI to a Senate that still calls people Nipponesians when no one is looking. We're doing the best we can alright, get off my back man."
We have a situation, but it's not exactly A&M Records, Inc. v. Napster.
Oh it is, and at least anthropic has paid $1.5 billion and deleted there torrented copies and not released any models derived from them as a consequence.
The thing is it turns out to be not that expensive to just buy a copy of every book legally and scan them. And there's even precedent that this is legal predating LLMs (Google books)
I have a bridge to sell you
Facts are in fact knowable, and the US legal system is in fact not terrible at getting to them.
But we've gone through some pretty weird times too. Turn of the last century was pretty tech billionaire edits, reconstruction was uh, not smooth, it's a mixed bag.
And most takes I hear seem to acknowledge that this is one of those weirder times: serious election fraud rhetoric from most everybody from 2016 to the present, very politicized courts (on both sides to be clear), very soft on anti-trust, very soft on adventurous accounting. The Epstein files and like, no consequences (pretty much uniquely for a developed nation with Epstein people). It's weird right now.
And I think I would be hard pressed to think of a weirder part of this weird time than the rule of law meets AI. We can haggle on where laws end and norms begin (stare decis being maybe the midpoint), but in the 90s, the Justice Department got their brass knuckles on for a lot less.
I don't think it's a simple "the law works nothing to see here" story.
I can understand why as someone who didn't follow it and the more corrupt legal developments closely you wouldn't be confident in that.
Granting people some form of control over knowledge only serves the public interest inasmuch it provides incentive to create more of it. Mass media, effortless duplication, and copyright extensions had already broken this to the point where control of knowledge was suppressing creation of new knowledge more than it facilitated.
The world has changed, we need a mechanism that works for the public interest that applies to the facts as they now are.
> The problem with distillation attacks
Apologies for repeating myself here, but what you call "distillation" is function approximation.
I feel for the teams at Anthropic and Open AI, but unlike startups from prior eras; Anthropic and OpenAI have decided to be in the business of selling compute. Not creating a product that uses compute, but a product that's math running on compute. This is different from what Google is (or, rather was. As always, RIP Google 1998-2019).
Google's algorithm might be math, but Google search isn't. Google search is a process that's continuously operating in the background. Google crawls pages. Google stores and indexes what it finds. Google then exposes this to retrieval via its algorithm. User uses algorithm.
Now, let's compare that to AI models. When Anthropic serves Mythos / Opus etc, they're taking input or x from their user, doing compute, and then serving the result of the Mythos / Opus function, i.e.,
f(text) -> (text_transform)
According to Stone-Weierstrass, given enough values of y for f(x), anyone can approximate this function.
The fidelity and sophistication of this approximation definitely requires a lot of cleverness and effort, and it is arguably an imposition on Anthropic and OpenAI. But on a long-enough timeline, they don't even have to poll Anthropic or OpenAI. As the internet is flooded by PRs, content, emails written by Mythos / Claude, and just people otherwise sharing the results of Claude prompts, then there's an ever increasing set of data to approximate the f(x) that's f_Claude.
Eventually, in the future, anyone will be able to create a good enough approximation of the f_Mythos. Which is Anthropic's product.
Anthropic and OpenAI can now wage war on mathematics and the open-ended compute. Or, they can adapt and build a better product.
Choosing Option B was the Silicon Valley option / choice. I think the OG large-scale Valley lobbying effort, the Semiconductor Industry Association, was unique in that it prioritized and chose to do real research.
https://en.wikipedia.org/wiki/Semiconductor_Industry_Associa...
https://en.wikipedia.org/wiki/Semiconductor_Research_Corpora...
This helped the industry to survive and outcompete the pressure they were facing (at the time).
These Chinese labs are producing novel models, publishing their techniques and sharing their open weights and the first topic of conversation is how they stole from U.S. AI labs.
Setting aside the fact that it doesn’t make any feasible sense to do API distillation, these models are outperforming frontier models on a number of benchmarks, and often times run more efficiently by several orders of magnitude.
We have to stop crying distillation, it’s getting embarrassing and at this point feels even a bit delusional.
Anthropic stated in February that Moonshot AI (the creator of Kimi) distilled ~3.4 million exchanges from Claude models, as explained in their press release https://www.anthropic.com/news/detecting-and-preventing-dist...
DeepSeek and others like Minimax are publishing deep research on Multi-Head Latent Attention and Mixture of Experts, Multi-Token Prediction, novel Sparse Attention approaches, I mean they trained long context models on a fraction of the resources and gave everyone the recipe.
Chinese labs might not have the funding of labs like Anthropic, but at least they provide the receipts.
This behavior is exactly what you'd expect from a model distilled from Claude.
Someone even took the time to analyze Kimi's ambiguous identity, in great detail: https://github.com/rgreenblatt/which_claude_is_k3/blob/main/...
And there's an entire Reddit thread discussing this https://www.reddit.com/r/LocalLLaMA/comments/1m2w5ge/did_kim...
That doesn’t prove Anthropic’s specific 3.4m-session allegation, but calling it “zero evidence” is no longer credible.
Kimi K2.5 was worse in a hilarious way, it identified itself as Claude and referenced Anthropic's Constitutional AI as some of its guiding principles https://huggingface.co/moonshotai/Kimi-K2.5/discussions/38
This is not at all what I would expect because it's trivial to change the training data to replace Claude with Kimi. In fact I'd argue it's almost certainly not saying that due to distillation.
K3 reproduces Claude's internal model identifier when prompted, something which the real Claude models themselves do not emit. This is highly suggestive that K3 was trained on Claude metadata (API logs, tagged synthetic data), rather than Claude's chat outputs.
And it's well documented that Chinese labs are buying large amounts of raw Claude metadata https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
"Caveat: fully AI-generated research."
And that you quoted or paraphrased directly.
I don’t consider a tweet by Denise Wu, who works at Anthropic, to be reproducible evidence.
I don’t consider “Caveat: fully AI-generated research.” To be someone taking time to analyze anything in great detail.
Because two AI models produce vaguely similar front-end styles when generating similar prompts I also do not consider to be of much value?
I think this is what I mean when I say the U.S. has its head in the sand. The Chinese labs are releasing ~60 page research reports with citations and analyses and evidence and Anthropic is throwing up defensive blog posts with zilch. I’ve seen more detail in a tech blog from Uber than anything I’ve seen from Anthropic.
"Zero evidence" as you claimed earlier isn't accurate. You've moved the goalposts from "evidence" to "raw internal logs I can independently audit," which is a different and very high standard. Sure Anthropic didn't publish logs, IP addresses, timestamps, or account IDs of the accounts involved. But that's true of any cybersecurity breach/abuse disclosure ever made. Companies are furtive to reveal how they detect fraud, because doing so exposes the signals used to detect bad actors, and makes future abuse easier. Not revealing the "evidence" you're asking for is industry standard practice. You're complaining that Anthropic is following industry standard practice, and conveniently defining the "evidence" you need as something Anthropic is never going to publish.
> I don’t consider a tweet by Denise Wu, who works at Anthropic, to be reproducible evidence.
Is the issue here that she works at Anthropic? Because Denise Wu doesn't work there.
> I don’t consider “Caveat: fully AI-generated research” to be someone taking time to analyze anything in great detail.
The experiments were run by Ryan Greenblatt, who is a real AI safety researcher (at Redwood Research).
The identity experiments and Greenblatt analysis are trivially reproducible. The methodology, code, and metrics are all there in the Github repository. You can ask your preferred AI to independently replicate these results, and it will give you a result within an hour.
You’ve also reduced the evidence to “two models producing vaguely similar front-end styles,” which is not what either analysis shows.
From the analysis, Kimi K3 identifies itself as Claude 15% of the time. How do you explain that? Qwen and GPT identify themselves as Claude 0% of the time.
If a long document is too much analysis for you, someone else made a simple chart which measures the KL divergence between Kimi K3 and other major models. They found K3 is unusually similar to Fable 5 & Opus models. That is, Kimi K3 has an very similar style and phrasing to that of Anthropic models. That behavior is expected from a model distilled from Claude.
Assuming each session was 10,000 words each, that's 34 billion words; lets call it 50 billion tokens (0.05 trillion) unfairly pilfered from Claude. That left Moonshot needing to scrounge for the other 14.950 trillion training tokens required for a baseline frontier model.
They are used for post-training, i.e. calibrating the model to understand and use tools/command line more effectively.
That's an increase of only a single order of magnitude, increasing my estimate of exfiltrated tokens from 0.05 to 0.15 trillion - a far cry from the 15 trillion required.
> They are used for post-training
Possibly - it may be too much data for post-training, unless further curation was done. However, this is not distillation; you know it, I know it, Dario knows it, but "Distillation Attack" is a short, memorable, sciencey-sounding, political sound-bite with enough malevolence to be deployed on the floors of congress, or by the usual fear-mongering newstainment talking heads.
Nobody is suggesting Moonshot used 15 trillion tokens of Claude data to pre-train a base model from scratch. That would be impossible and nonsensical.
This is entirely about distillation, which happens during post-training (alignment and SFT). Here, datasets are measured in millions or billions of tokens, not trillions. 50 billion Claude tokens is far, far than enough to copy Claude's reasoning logic, writing style, and tool-use ability to the pre-trained base model.
> However, this is not distillation
I don't understand how you're so caught up on the term "distillation". Distillation is using a larger model's outputs to train a (weaker) student model. Which is exactly what's happening. It's a standardized term that has been in use for a decade.
Now that Anthropic hides the real thinking tokens in a way that precludes future CoT distillation, we'll find out which side is correct based on whether Chinese AI labs close the gap or not.
My bet is they'll close the gap; nothing about frontier AI is magic, once something is shown to be possible, experienced practitioners almost always figure out how to accomplish the same feat, though not always on the same way. This is why frontier US labs keep leapfrogging each other every few months.
It's a PR campaign - when they say its an "attack" they don't mean on Anthropic - but on America itself. What kind of American can let such a brazen attack go unanswered? At the very least, they ought to demand the dangerous, pinko, stolen models be banned in all 50 states, and pay whatever price demanded by the patriotic, freedom-loving, all-American AI labs that can never be accused of stealing.
https://en.wikipedia.org/wiki/Feist_Publications,_Inc._v._Ru....
And then there's new updates related to AI that fully take out LLMs from protection.
What a nice post hoc revision of history. Distillation is still an active area of research, that you can distill models as easily as you can it genuinely interesting and absolutely not something that was taken for granted even 12 months ago.
Even 6 months ago this idea that 'using model outputs as training examples' was listed as the reason that all models would fail in the near future due to some spooky circular training catastrophe.
Don't pretend like this was so obvious.
yeah hardware companies make for nice stories or green numbers on Wall Street - but value will be captured by application layer.
look at history.
what is the end game for this strategy?
if the frontier labs shut down, or stop releasing to the public, and there's noting left to distill, how will you progress?
I'm pretty sure that all labs are distilling each others' LLMs, maybe apart from Anthropic and OpenAI. It would be stupid not to do it, because it's cheap and effective. But that's not the only thing they're doing. If you think K3 and GLM-5.2 got this good only from distilling frontier models, you're not paying attention to Chinese labs' publications.
if this were not the case, then we would be observing chinese models that far surpass frontier models in capabilities, rather than "almost as good, but much cheaper", and we would be having a very different conversation. what happens to these efforts when the subsidy is cut off?
I see no evidence for that.
> if this were not the case, then we would be observing chinese models that far surpass frontier models
It's pretty clear that the primary reason for the difference is budget and compute availability. Chinese labs have at least an order of magnitude less money than Anthropic and OpenAI.
> what happens to these efforts when the subsidy is cut off?
They will continue making progress as they do now, minus the benefits of distillation.
Moonshot AI Scale: Over 3.4 million exchanges
The operation targeted:
Agentic reasoning and tool use Coding and data analysis Computer-use agent development Computer vision Moonshot (Kimi models) employed hundreds of fraudulent accounts spanning multiple access pathways. Varied account types made the campaign harder to detect as a coordinated operation. We attributed the campaign through request metadata, which matched the public profiles of senior Moonshot staff. In a later phase, Moonshot used a more targeted approach, attempting to extract and reconstruct Claude’s reasoning traces.
Translation: we have the machinery in place to identify our users, and actively do so.
Once a model becomes competent enough to perform complex reasoning, a teacher model is no longer necessary. The model can now reason about its own behavior and build a better version of itself through recursive self-improvement (RSI).
Kimi K3 is capable of RSI.
and what will fund these budgets exactly? inference is cheap, distillation is cheap, training is what's expensive.
Say it louder for the people in the back. All these complaints about "distillation" from frontier labs are bordering on felony contempt of business model at this point. It's great for us. Maybe it's bad for them but nobody other than shareholders really cares.
The optimal outcome for humanity is for oligarchs to spend trillions training a godlike AI, only for the precious weights to just leak. No "distillation" required.
If "distillation attacks" happen, we have to conclude there is some value add in what model labs do. Regardless of how we feel about using existing human knowledge in the way they currently do, it's simply impractical to infer that everything that happens downstream of LLMs can not be an attack on some IP because of it.
So both things can be true: a) People infringe on Anthropics IP and b) what Anthropic did to build their models is legally questionable (or might be ruled illegal, even though I doubt it).
No.
Authors do not infringe on IP when they read another's book, nor should the lumber company be able to dictate how I use planks and if I can resell them if i'm done with them.
You're framing it as if the added value of the author or lumber company, awards them consideration when somebody uses the products to create more value.
IP law was always a big mess, and these questions cross far into ideology instead of law; but I do not understand people who think we need an ideology where more IP-law is good for society.
Are the distillers reading books or are they building models?
If anthropic is providing no value they can just build from scratch. But obviously distilling is easier. Hes saying thats the value they add.
This would demolish agent usage by corporations.
Unless someone literally stole the weights somehow (which is not out of the question, I doubt either oAI/Anthropic have the capabilities to prevent a state-level actor getting those weights), distillation from generations is not infringement on anyone's IP nor is it stealing nor is it an attack. It can't be. As long as you pay for tokens you get to do whatever you want with them. Someone saying you can't doesn't mean it's an attack or their IP or whatever. They either sell the tokens or not. They can decide to not sell them to anyone, but again that's not stealing.
And their ToS are a joke. Imagine how people would react if MS had ToS saying that you can't use MS software to develop solutions that compete with MS. They'd be laughed out of the room. Somehow it's ok for token sellers to decide what you do with the tokens? Why? If you pay for something you get to do whatever you want with that output. Train, distill, whatever.
I guess I can see that, if you mean the targeted effort of creating many accounts w/ the intent of doing it at scale. Sure, they may see that as an attack. But again, it's only an attack from their perspective if you agree that using generations to distill is "wrong". I just don't see it, in general. You can't both sell tokens and decide that distilling is somehow illegal. Something, something, cake and eat it.
How do things get "established" from someone's perspective, exactly?
By that logic it is established from my perspective that Anthropic has no right to train on anything I've written that is publicly available on the internet.
Of course, they don't care about my perspective, but then again I don't care about theirs.
Anthropic’s model outputs contain no IP. This is actually a simple legal proposition (rare in this field!) that derives from the fact that only specific classes of IP exist: copyrights, patents, trade secrets, and trademarks. Examining each, it is clear that API outputs do not qualify. Anthropic disclaims copyright in outputs; the outputs are not patented; the outputs are not secret (a prerequisite to having trade secrets); and trademarks are irrelevant in concept.
Anthropic, OpenAI, etc do not deserve legal protection.
that would be existential doom for them because then they have a case to claim ownership of their users' codebases
no corporation would sign off on that
Do you really think intellectual property laws will prevent this in practice? It’s like as if we said, “hey, USSR, you can’t make a nuke, too! We patented that already.”
Asking China to not distill our models down is equally as ridiculous.
How far are we willing to go as a nation (and as a species) to prove out the scaling laws? Are we willing to sacrifice our industrial base? Would we rather train models or smelt aluminum?
(I actually appreciated your analogy, despite my lark)
So why didnt we have these LLMs in 2005?
Don't forget to account for all the costs. It's not just that CPUs are X times slower. Memory is X times smaller, too, and networks are X time slower. And all this hardware is many times more expensive.
If I'm getting my mental estimation right, training a 2026-frontier-class LLM in 2005 would be somewhere on the order of all the computation power in the world at the time. It's not that many more factors of magnitude before you end up at "all the computation power in the world up to that point".
"Distillation" literally means to separate and take some components out of something. You can distill how a model works from a model. You cant distill a model from information because the information does not contain the model.
People are happy to conflate distilling with building because they dont like how the information was used. You distill how the model works from the model, and you build a model with information. Both could be morally good or bad but its not the same thing.
Not really, what the information actually is, matters a great deal. It's harder to get good results going from "nothing > model+weights" than "nothing + traces from known good sessions of other good model > model+weights", this is what the "distillation" part is referring to. If "information is information", you wouldn't even need to separate good from bad sessions while doing the training, which leads to somewhat obvious results if you don't.
To succinctly restate my point, you cannot distill a model from information because the model is not contained within that information. You can distill a model from another model.
1: That's the "T" in GPT fyi, even though Google is the author of the research paper that changed everything
So the initial models arent just distilled from information. We’ve always had the information.
The first "L" in LLM does the work. In 2005 you had no Github, Stackoverflow, Youtube, common crawl and no archive of digital ebooks.
Once western governments declare it to be a "national security" risk for citizens to have access to open-weight frontier models, and once they classify using these models as acts of terrorism, what will that world be like?
Will using Kimi K3 come to be like how napster was in the olden days? Everybody knew it was technically illegal, but come on -- any track at your fingertips? But surveillance is quite more evolved now.
Or it will be like cannabis, where a guy in the neighborhood will low key rent you metered access to the 8x5090 rig in his basement he cobbled together from parts on ebay? Or everyone will flock to VPNs?
Or will the oppressors actually succeed? The same way that napster is long gone, and everyone accepts that they must pay spotify for a homogenized collection, where artists must take only a minuscule cut (more than napster though)... We'll be stuck with nerfed Cohere or Mistral models for open-weight options, as if they need more lobotomizing. Or else we can pay through the nose for Anthropic/OpenAI for "American Frontier" models which will fall increasingly far behind China.
Or else, like how Kindle Fire was subsidized by ads, we'll have "Kindle AI" where influence is sold to the highest bidder, where the LLM will tell us that smoking is actually healthy if big tobacco can engineer its renaissance by turning its lobbying dollars to pay-to-play, pumping its propaganda into the training pipeline for Amazon's extra commercialized line of ultra budget LLMs.
So maybe some isolated switzrland/singapore type locales would exist for US/EUusers to be able to dip their toes across the curtain legally without reprucursions.
[1] https://nitter.net/RnaudBertrand/status/2069574934972797089
If you need infrastructure done, China is dominating that area too. Rail, High-speed rail, Nuclear reactors, (near future Thorium reactors), Dams, Highway roads, bridges, Ocean ports, airports you name it, and they can roll it out, Transport ships, And if they don’t do it, Japan, Korea, Vietnam, and Taiwan do.
Is it too late? No, not necessarily, but America needs a regime change…
America is what it is. The only thing that will change it is leaning in not bemoaning "rural people" on Hacker News.
Are you talking about the US, specifically?
Why would other countries, that don't share the same anxiety about China as the US, would be troubled with the this?
It's the other way around.
There is a high likelihood that many countries of the "west" (the "global north"?) will outlaw, restrict, or otherwise control LLMs and the tools that enable them.
The US, however, is blessed with the first amendment which makes it extremely difficult to restrain speech in any form - including code.
It's not just the tariffs and imperialist/autocratic aspirations of the current President; it's also the fecklessness of the federal legislature and the revelation via social media that a large cohort of the public hold a negative-sum worldview and enthusiastically endorse bad faith dealing.
It’s going to be a different world, a world where many former allies are not gonna look to the United States first they can no longer afford to.
It has nothing to do with running open models, especially in hardware within Europe.
I actually have this trophy from the previous bursting bubble ... a Sun microsystems rack populated with three e4500.
$750k + of equipment at original list price ...
I only have the $20 plan from OpenAI and the same task, with a lot of the same implementation details as Kimi Code, only took a few minutes and consumed almost none of the 5 hour limit.
Subscription usage limits are hard to measure as none of the providers tell you directly what it means in terms of tokens or anything else you can easily compare, but when I sat down to add Kimi Code to flar, it was because I wanted to try it on some real work and then couldn't do any, because usage was nearly gone after the trivial task...no other ~$20 subscription I have has felt that tight before.
So, it was really slow to complete the task and seemingly much more expensive than every other model I'd tried. Maybe bad luck. Maybe it'll do better on other tasks. I wouldn't know as I was out of usage when I had time to try.
It did find a bug that Gemini 3.5 Flash introduced unprompted, though, so it has that going for it.
But I agree that price per token figure is not great. It seems even the tokens per character can vary between models, so it's basically useless.
In my own experience, Fable is more token efficient than opus 4.8 with a higher likelihood of completing tasks correctly or at least with minimal corrective work. Opus regularly struggled to gather the correct context and reason effectively about what it had gathered.
GPT-5.6-sol crushes fable in speed and token efficiency and is clearly superior across many tasks that matter for me.
I also find all models from anthropic after opus 4.6 to suffer from the same ai slop language that long plagued OpenAI and seems to have been reduced drastically in 5.6
I gave Claude Code/Fable the same task and it took significantly less time, but also stumbled on the same error as GLM. I didn't have it fix it though. I was mostly interested in timing differences.
I do like open models where I can, but I'm really hoping they get trained to second guess less. Or maybe I just need to prompt them differently. I'm not sure.
Then I typed /code-review in a second terminal/clean session after the analysis was done (no code changes) the usage was 99%. I then asked it to write that into a review.md so I could restart from that the next day. Sadly the last % wasn't enough for that.
Ymmv, these models behave very differently with no discernable reason. Usually reviews(even with fable) take like 10-20%... Yet suddenly you get it to burn through 65-69% in 15 minutes or so
AI subscription pricing is so goofy. You get some amount of usage that varies by models, is measured by opaque token usage, driven by how many tokens the (usually) vendor-provided interface (or model itself) wants to use. Then your usage is limited by time opaque time windows.
Now, of course, the plan is to remove Fable from the subscription. To paraphrase Darth Vader, they have altered the deal. Pray they do not alter it further.
What OpenAI in particular have done with reasoning efficiency in the past few months since ChatGPT 5.5 is nothing short of remarkable. It's overshadowed a bit by the benchmark game and the Fable hoopla.
Now is the time to focus less on token cost and intelligence, but tokens to solve a particular set of tasks in closed benchmarks for a variety of categories.
What is the use of grand intelligence if it either costs you a kidney or can't complete at all within a token budget? Even if there are niche uses where you truly want "maximum power" above all, we need to at least more severely penalize such models versus those that does it just as fine within a tenth of the token cost.
I'm aware of some benchmarks at the Artificial Intelligence site, but CLEARLY we are not focusing enough on these today and still leaving the fun surprises to the users.
When I looked at traces from benchmarking, I saw a lot of backtracking and uncertainty while reasoning ("wait, but..."). This also happens with GPT 5.6 and Fable with xhigh/max thinking, albeit to a lesser degree.
I think that explains part of the token inefficiency. Hopefully it will improve with lower reasoning effort settings.
[1]: https://platform.kimi.ai/docs/guide/use-thinking-effort
... but I borrowed a friend's +86 phone number, which you'll need to even see that price. or maybe a 回国 VPN will work.
And, DeepSeek is what I use for any task that works best with an API. It's cheap enough to where I don't think about cost, made even cheaper by DeepSeek having the most effective and cheap caching in the industry, and it's good enough to where I rarely have to follow up with a more expensive model or manually fix things. It's been alleged they're releasing an update to DeepSeek V4 Pro soon that improves it, which likely makes it a good fit for even more kinds of problems. It remains my favorite of the Chinese models, it's so cheap and cheerful. And, is also less aggressively censored than some of them.
I use DeepSeek v4 Flash & MiMo v2.5 Pro. Prefer the latter over DeepSeek v4 Pro because it costs the same while being equally good & less chattier for coding workloads. Although, I've begun experimenting with Hy3 (as an in-between Flash & Pro) & GLM 5.2 (for long-horizon tasks).
Subsidies would affect 1, but not 2. But if some VC wants to subsidize my Claude or Codex or whatever, awesome.
Additionally that same VC could be (read: is always) spent on developing the harness, and other infrastructure around the model, not just the model itself.
So it's apples-to-oranges when comparing a relatively new model to established competitors (i.e. OpenAI @ $900B funding vs Moonshot/Kimi's $30B FYI) because every new model they release is judged on "performance" which is not strictly speaking derived solely from the model.
It's possible Moonshot could get similar performance over time as the build out the rest of the infrastructure. We have no way of knowing how much of OpenAI/Anthropic's success is due to the model vs intelligent tooling built on top of it.
However I've seen some benchmarks say it uses fewer than fable which hasn't been my experience.
ArtificialAnalysis puts Kimi K3 just below DeepSeek v4 & GLM 5.2 in token use per task, which is about 2x to 3x more tokens than Grok 4.5: https://x.com/ArtificialAnlys/status/2077832879187620192 / https://archive.vn/zBbFi 2 other open weights MiMo v2.5 & MiniMax M3 are comparatively thrifty.
> Subscription usage limits are hard to measure as none of the providers tell you directly what it means in terms of tokens or anything else you can easily compare
I always put my coding subscriptions (that allow it) through "AI gateways" (Cloudflare & OpenRouter are free) which help track token use.
In my experience, Kimi & Qwen Cloud have opaque & restrictive limits, their "credits" drain faster. I now make it a point of subscribing (directly [0]) with providers that are transparent like MiniMax, DeepSeek, Xiaomi, & Z.ai.
[0] OpenCode Go, Cline, and AtlasCloud have generous limits for open weights, otherwise.
The benchmarks come out and say they’re as good as Opus from N months ago, then I use it for a complex task and it doesn’t work as well as Opus from N months ago did when on similar problems.
There’s a real wow factor when you get an open weights model to do amazing things, but in my experience the gap to the frontier models has always been bigger than the benchmarks would lead me to believe.
There can be a lot of value in having the cheaper open weight models for chewing through lower complexity tasks (non-programming in my primary use case) at a cheaper rate than OpenAI or other frontier API costs. Even with those I can measure bigger gaps to the frontier models than the benchmarks suggest.
If the benchmarks aren’t being directly gamed, there’s at least some selection happening where training data or model structures are being picked in ways to maximize public benchmark performance. All of the labs know there’s immense value in having good benchmarks to show for your model because most LLM consumers are picking based on lab provided benchmark charts, not running their own evals. Running your own evals is hard and expensive.
This does depend heavily on the kind of work you do and how you use these models, but the idea that K3 isn't right up there with US SOTA models doesn't match my experience.
That's weeks maybe months behind, not months maybe a year behind. It's "would my life really change if Claude was gone, not really" behind.
I actually haven't used it much, because Claude started kicking ass again the last few days. Like, way too much of a difference to be normal load-based variance. I got more done in the last 48 hours than week before that.
So, fuck yeah competition.
Kimi K3 has 2.8 trillion parameters. We don't know the number of parameters of ChatGPT 5.6 or Opus 4.8, but it's probably in the same region. Fable/Mythos are rumored to be around 10 trillion.
So, K3 is directly comparable with ChatGPT 5.6 and Opus 4.8, and the price is not so much lower:
K3: $3/$15 per 1 Mtok input/output ChatGPT 5.6 Sol: $5/$30 Opus 4.8: $5/$25
This is not a watershed moment. It's a competitor converging to the same capability and trying to undercut your prices, but not by a lot.
As for the open weights? For now, Kimi K3's weights are closed, and I don't expect the situation would change.
It'll change on July 27 (based on https://www.kimi.com/blog/kimi-k3):
> The full model weights will be released by July 27, 2026
When you say "Claude", do you mean Opus? Fable? What effort level?
Those things do make a difference to some of us, even though nothing is black and white. In my case, I'll probably want to wait until other providers appear through OpenRouter and then I'll try to judge how much I trust them. But even if I don't trust them much, they don't train models anyway, so the likelihood of my data being used that way is smaller.
I find these kinds of concerns increasingly silly: most of the input to these models will be ... previous output from the very same models, alongside the occasional half-assed human command to fix something and "make zero mistakes". Who cares if they train on that? Let them, if it makes their future models better!
99% of users are not working on any special IP to worry about that.
ref: https://www.kimi.com/code/docs/en/kimi-code/models.html
If you sign up with non-Chinese phone number, you're bucketed into US, you get US prices, can pay only in USD and with American credit card network.
Chinese prices are about 9x cheaper than the US prices, which are already far cheaper than Claude or other American provider. If you can somehow get hold of a Chinese phone number, keep in mind that you can save ~90% of the bill.
My assumption is that Anthropic, OpenAI, Kimi, etc all have a similar cost structure when serving models. The same size model roughly generates the same GPU usage whether you’re American or Chinese. I’d also guess that the model sizes across all SOTA models is similar, we just only see data for open models. The difference is most likely that American companies simply charge more because they have the dominant market position.
Remember not too long ago when Anthropic was charging $75/mt for Opus? Now that many models are in “opus tier”, their pricing is $25 - higher than competitors but close. The newest Kimi is $15. 40% lower to forgo “made in America” with American enterprise support staff is not crazy. Compare AWS to Hetzner or any other flagship enterprise service to the foreign and discount option. I assume that over time, we’ll see the commodification of models reducing prices even towards the raw GPU costs.
That may have been closer to reality 10-20 years ago, China is a different country now, what I mean by that is they offer research funding, they have huge digital behemoths (alibaba, tencent, huawei, bytedance etc), large scale deployment opportunities and prestigious careers. Many graduates return because the opportunity set is attractive and they want to return, it's not just because US immigration policy pushed them out. Some also want to contribute to their own country's technological progress (which is a normal motivation btw), like probably you are also a patriot and want your country to succeed.
So, really, China's AI progress is not mainly the result of America failing to absorb every talented Chinese researcher. China has built a domestic ecosystem capable of producing and keeping top talent itself. I feel like a lot of Americans do not understand this.
[1] https://www.latimes.com/world-nation/story/2025-02-21/why-ch...
[2] https://www.wsj.com/world/china/americas-allure-fades-in-chi...
[3] https://www.theguardian.com/world/2025/jun/06/chinese-studen...
US immigration policy may be unnecessarily pushing away talent but the assumption that talented Chinese researchers would naturally remain in America unless prevented from doing so ignores the growth of Chinese universities/labs, companies, their funding, national prestige etc.
I mean, don't get me wrong, US is still highly attractive, it is just no longer the only place where an ambitious Chinese researcher can do important work and grow.
20k O-1 visas were issued last FY which was mostly under the Trump admin, up from 19.5k the previous FY under the Biden admin
When people demonstrate their capability thoroughly, the Chinese government takes away their passports. You’re not exactly going to get them here with an O-1.
Basically of all visas O-1 is virtually guaranteed to have highly positive economic value
US immigration policy isn't a big factor.
China's got 1.8B people. If you don't think they've got the talent to pull this off, even if a lot of it leaves to live elsewhere, you're naive.
No one uses Baidu, but they built their own Google, and it's good.
They built their own Facebooks and Instagrams.
The US isn't the only place in the world where people can build software...
- china’s homegrown tech industries already achieved escape velocity from it a long time ago, after China fenced off its market for Alibaba and Baidu in the ‘00s. some of their AI innovation at the edges was already top class 10 years ago
For example, Kimi 2.7 has been really effective for me despite having verbose thinking blocks, simply because it runs so fast. Speed-wise, it feels about like Sonnet, possibly faster.
Can’t use for commercial purposes. Can’t opt out of training. Data retained.
"Can't use for commercial purposes" - incorrect AFAICT. In what sense do you mean this? The open weight MIT version obviously allows for commercial use, but I don't think that's what you're referring to, because training data is irrelevant on the open weight version. Pretty sure the API allows commercial use too. Maybe the free version doesn't? But who cares?
If the author is here, I'm curious what this means. How are they running Kimi K3? Are they using pi, opencode, claude, codex, or kimi-cli? Is speed a concern?
Without knowing how the comparisons are being made, it's hard to agree that one can't notice the difference. I do.
I don't understand how a product that:
- is interfaced with and is deeply linked to natural language, so everything you produce (sessions, history, etc) is in Markdown and you can literally install a second model and tell it "hey import all of Claude's memory into yours" and that's it
- is based on well understood technology, the real constraints are how much money you put into training the models, but the theory has all been developed in the open
- clearly has a threshold where it quickly commoditises and turns from "I want the best" to "hey the best is a bit too expensive. The second best is half the price and works close enough".
was ever supposed to be a money printing machine. The fact something is extremely useful doesn't imply it's extremely profitable.
IMHO we're clearly speedrunning the process of turning AI into a commodity. Dario Amodei knows pretty well that when or if Anthropic cuts people off Fable, the vast majority of them will definitely not pay for it because Opus 4.8 is good enough for almost everybody that _knows_ what they're doing, and so are basically half of the most recent models. If I already have good baking skills I don't become more productive with an automatic bread machine, I just need a better dough mixer and oven
A closely related question is “what do the American labs need to do in order to justify their enormous market valuations?”
It seems like the answer cannot possibly be “gradually improve model capability while figuring out how to better monetize inference.” The valuations are just way too high for that to be sufficient.
Surely the answer has to be “continually achieve large leaps in capability comparable to the first consumer releases of ChatGPT while also maintaining a significant capability lead over open models and new competitors.”
And does anyone think that’s going to happen? Even with state-level protection from competition (which incidentally would significantly harm the American economy), the large leaps in capability seem to be coming fewer and farther between.
What appeared initially to be a huge innovation was later easily duplicated by many. There are no platform-lockins or network effects. Switching costs for users are zero, and there are low barriers to entry, with vast numbers of models to choose from and more appearing every day. As a business a token will be a commodity like an electron. Doesnt matter who produces it, or how (solar, wind, coal, nuclear etc) as long as it powers my toaster.
everything else we see today is just preparing for it.
It's easy to prove.
Serve your model to billions of users per month.
1) Kimi 3 is a "very good model"
2) It's performance can NOT be explained by distillation
3) The US government should create FUD to stop US corporations from using it (so they use OpenAI instead)
"One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a 'public good' which will ultimately be provided by the state as a kind of 'digital public infrastructure.' This future strikes me as a dystopian hellscape, but I've never met an open-weight models advocate who doesn't ultimately concede this is where things end."
He never says why he thinks AI as a "public good" is dystopian, but it's not hard to imagine why. It's because he and his inner circle won't have the power to dictate what we read, see and hear.
Is anyone using open source models for anything major ?
What is the parento frontier?
The Pareto frontier tells you which designs are the best in at least one of your metrics (non-dominated by another design). For example if you're selecting a car and you care about both speed and mpg, a Formula 1 car and a Prius might lie on the Pareto frontier, but a Model T Ford would not.
So like, on a cost-intelligence graph, the cheapest and most intelligent models are pareto optimal. Then in-between those if you have
- cost $3 intelligence 6
- cost $1 intelligence 5
- cost $2 intelligence 4
The 1st and 2nd are pareto optimal, the 3rd is not, because it's dominated by the 2nd (2nd is cheaper AND more intelligent at the same time)
On Openrouter Kimi K3 says it does not retain data or train on it, which is better than what US hosts claim for Claude, ChatGPT, etc.. as they collect and retain data even if you disable training on it.
Opencode or similar open source tool + a zero data retention provider is about the best option aside from running a smaller fully local model on your own PC.
Here's the thing about this though, the auto industry directly employed hundreds of thousands of people.
The AI labs are small, only few benefit directly from their wealth and there's already immense opposition to AI, data centers, etc...
Kimi K3: Open Frontier Intelligence
https://news.ycombinator.com/item?id=48935342
Kimi K3, and what we can still learn from the pelican benchmark
"...I’ve been running Kimi K3 alongside Claude on my normal coding work, and for all practical purposes I can’t tell them apart. Same tasks, same quality of output, and near identical token counts to get there. I expected an open model to be sloppier or to grind through more tokens on the way to the same answer, and neither turned out to be true.
The prices are nowhere near each other. K3’s API runs $3 per million input tokens and $15 per million output. Claude’s top model costs $10 and $50 for the same units. The subscription side is even more lopsided..."
> When the headline model on your plan can be switched off because the economics don’t work, the plan was never really selling you the headline model. Kimi’s tiers don’t come with that asterisk.
This line has a certain smug, punchy cleverness that I associate with AI. To me, the vibes are ~30% AI writing.
I didn't.
In a few years there will be Mythos level open weight models hosted by the lowest bidder anyway.
At the rate things are moving I'd expect that to happen much sooner.
In fact: Somebody, right now as we speak, is most likely already working on training the next best open source model.
I just thought about that recently too, then Kimi K3 came out, and I thought: Yea, I'm not surprised. Just a matter of time now...
And this is the point where your internal compiler should have started shouting 'Type Error'
Notice the trick here?
> Then there’s the fine print. Claude couldn’t sustain Fable access on the twenty dollar plan, so they turned it off, and the plan quietly falls back to Opus.
Where is the Fable-class Kimi model at all?
i posted the transcript as a reply to you, and HN automatically flagged it.
but go on.