Kimi-K3 Technical Report [pdf]
Posted by vinhnx 2 hours ago
Related: Kimi-K3 on HuggingFace - https://news.ycombinator.com/item?id=49065752
Comments
Comment by GodelNumbering 1 hour ago
If you are a large enough company that spends million+ on inference a month, it makes sense to buy a GB300 rack ($6M on top range from what I could find) which has 20.7 TB. Since the model is mixed trained (MXFP4), you would need less than 10% of the rack's memory to serve the full model. Aggregate HBM bandwidth: 576 TB/s. You can run over 6000 parallel agentic workflows (each with ~100k context on average) at ~30 tok/s.
Assuming the annual amortization+electricity at $1.5M/year and about 50% average annual utilization, you get less than 60 cents (USD) per million output token, for a frontier model with plenty of capacity to share, all your data never leaving premises and well over an order of magnitude cheaper!
As long as a company believes that the openweight models will continue to get more capable and 'AI is here to stay', this model provides the first solid footing for a decision to just buy a rack.
Comment by 999900000999 53 minutes ago
That another 400 to 700k.
It becomes your problem and not someone else’s. However, I don’t trust hosted LLMs for anything that needs to be private.
Comment by wongarsu 48 minutes ago
Comment by arjie 45 minutes ago
Comment by LeonM 26 minutes ago
40% overhead is quite typical, so you'd be looking at $120k/year. In the USA I'd consider that a competitive salary for an admin capable or keeping a $6M rack of specialized hardware running 24/7.
Comment by russell_h 51 minutes ago
Why not? Do you trust AWS with things that need to be private?
Comment by Kevcmk 45 minutes ago
Comment by senderista 38 minutes ago
Comment by lumost 45 minutes ago
Having these models in the open caps the inference margin.
Comment by GodelNumbering 47 minutes ago
Not a devops but I'd say one full time is already too many.
Comment by slicktux 39 minutes ago
Comment by toomuchtodo 38 minutes ago
(managed an entire data center building with thousands of servers a lifetime ago with ~2-3 other people, it’s only gotten easier over the last two decades imho)
Comment by clint 51 minutes ago
Comment by reckless 53 minutes ago
Comment by petu 42 minutes ago
> 4. The requirements set forth in Sections 2 and 3 do not apply to: (a) internal use of the Software, defined as any use that does not make the Software, its outputs, or its underlying capabilities available to third parties; [...]
Comment by jsnell 14 minutes ago
Comment by vidarh 46 minutes ago
Comment by m_ke 2 hours ago
Anyone who claims open source and open weights models are "decel" needs to get their head checked
https://github.com/MoonshotAI/MoonEP
Comment by jvanderbot 1 hour ago
Comment by ike_a 1 hour ago
Decel:
- Potentially reduces investor appetite for funding big labs.
- More risk of powerful AI getting in bad hands -> more regulation.
Accel:
- More competition so big labs can't rest on laurels.
- More research in open, so all labs can accrete advancements faster.
I feel like open-source = acceleration has a much more clear argument. (and how bad would deceleration be in any case?)
Comment by sosodev 1 hour ago
The problem with the decel/accel rhetoric is that it lacks nuance.
Comment by Smaug123 1 hour ago
Comment by reissbaker 45 minutes ago
Comment by chorizo 26 minutes ago
Comment by StevenWaterman 1 hour ago
Comment by f311a 1 hour ago
I think you meant less research and experiments in big labs because they don't get all the AI money.
Training is expensive, but they also have more than 10 000 of employees combined and they cost a lot of money.
Comment by Iolaum 1 hour ago
Comment by zozbot234 1 hour ago
Kimi K3 is plausibly a lot less dangerous than a totally jailbroken ChatGPT/Gemini/Claude Sonnet (let alone Opus or Fable!) and it's quite deeply weird how no one seems to be calling for those models to be banned or restrained by further regulation. Why the double standard against the less concerning (but more efficient!) open weight models?
Comment by ike_a 1 hour ago
Is there a world where open source models end up at the frontier, or do you think there are structural/first-principles reasons why this won't happen?
Comment by zozbot234 1 hour ago
Comment by SirLordBoss 59 minutes ago
Comment by acedTrex 1 hour ago
Comment by viccis 1 hour ago
Comment by jvanderbot 10 minutes ago
Comment by Almondsetat 1 hour ago
Comment by viccis 38 minutes ago
Comment by SubiculumCode 42 minutes ago
Comment by Der_Einzige 1 hour ago
Comment by vanuatu 58 minutes ago
the only reason other labs can catch up is because the frontier labs can be distilled, and they siphon a % of the labs' revenue to reinvest into the next iteration
full accel would mean nationalizing the big 2 labs and locking in manhattan project style until RSI
Comment by m_ke 48 minutes ago
There's no chance K3 is a distill of Fable, it came out way too soon after the limited fable release to be feasbile.
If you look at all of the top ML conferences, chinese labs contribute way more to advances in ML than "Open"AI and Anthropic: https://www.reddit.com/r/TheMachineGod/comments/1pi4q7f/pape...
This K3 release just helped every other lab on the planet stay in the race by making it possible for them to build on top of it, placing them at the frontier starting line instead of having to spend billions of their own dollars and risking it all to attempt to catch up.
The open source contributions I linked to above will move the whole field forward and reduce the costs of training and inference for everyone.
Open science compounds on it self, every new advancement pushes the field forwards and opens up new grounds for future improvements.
It is impossible for a single closed lab to consistently stay ahead of the rest of the field, especially in a huge growing research area like machine learning. The only only advantage the big labs have is money, but the naive scaling game is not sustainable long term when you have to pay 10-100x more then the fast followers and we start getting more and more open models or use case specific models that can handle 90% of high volume use cases.
Research is a high variance, low expected value activity, meaning that the few large concentrated labs have to be conservative with their bets and double down on proven things when scaling up. The rest of the field is like a diversified portfolio, with thousands of players making smaller riskier bets that only require a few of them to succeed (like K3 did here, and DeepSeek a year ago)
EDIT: also if you look at most of the work from OpenAI, it's mostly taking existing promising open research work and scaling it up. (except for things like CLIP and etc from Alec Radford)
Comment by chorizo 32 minutes ago
And years down the line, lots of other research labs used my code and cited my paper.
Comment by m_ke 24 minutes ago
We had a bunch of things that we never published that ended up being major research findings years later at top conferences.
Comment by calebkaiser 13 minutes ago
Further, you can just read the papers released alongside most open models. Plenty of hugely influential research results published that drive the frontier forward. It's not like these models are just existing architectures downloaded from Huggingface and trained on frontier lab APIs.
Comment by ZeroGravitas 34 minutes ago
Comment by pluto_modadic 35 minutes ago
Frontier models would have to do something extraordinary or unique, or unreplicatable, because clearly there is no moat, and US companies are sitting on huge nvidia valuations and get surprised when competitors beat them.
Comment by fahrradflucht 1 hour ago
> If the Licensee or any of its affiliates operates a Model as a Service business, and the aggregate revenue of the Licensee and its affiliates exceeds 20 million US dollars (or the equivalent in other currencies) in total over any consecutive 12 months, the Licensee must enter into a separate agreement with Moonshot AI before using the Software or its derivative works for any commercial purpose.
+ the existing 100 million monthly active users, or more than 20 million US dollars for commercial products have to name Kimi clause
Comment by cosmojg 1 minute ago
Here's the relevant text[2] from the Compendium of U.S. Copyright Office Practices (3d ed. 2021) § 313.2, "Works That Lack Human Authorship":
> […] the Copyright Act protects “original works of authorship.” 17 U.S.C. § 102(a) (emphasis added). To qualify as a work of “authorship” a work must be created by a human being. See Burrow-Giles Lithographic Co., 111 U.S. at 58. Works that do not satisfy this requirement are not copyrightable.
> […] the Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author. The crucial question is “whether the ‘work’ is basically one of human authorship, with the computer [or other device] merely being an assisting instrument, or whether the traditional elements of authorship in the work (literary, artistic, or musical expression or elements of selection, arrangement, etc.) were actually conceived and executed not by man but by a machine.” U.S. COPYRIGHT OFFICE, REPORT TO THE LIBRARIAN OF CONGRESS BY THE REGISTER OF COPYRIGHTS 5 (1965).
[1] https://www.lewisrice.com/publications/d-c-circuit-affirms-h...
[2] https://www.copyright.gov/comp3/chap300/ch300-copyrightable-...
Comment by simonw 1 hour ago
The models they released in 2025 - https://huggingface.co/moonshotai/models - were clean MIT. They started doing the "modified MIT" thing in January 2026 with moonshotai/Kimi-K2-Thinking
Comment by zozbot234 1 hour ago
Comment by simonw 1 hour ago
Comment by lossolo 1 hour ago
Comment by simonw 44 minutes ago
"Modified MIT" is the perfect example of that.
I'm not saying it's unreasonable, or that you can't release under such a license - it's your software, use whatever license you like!
I'll call it "janky" when you do.
Comment by lossolo 31 minutes ago
Comment by simonw 28 minutes ago
Comment by bilbo0s 49 minutes ago
For people at the top of the tech pyramid, I could totally see why they'd want to put an end to all this "Open Source LLM" stuff.
Comment by throwaway27448 1 hour ago
Comment by paxys 1 hour ago
Comment by mike_hearn 12 minutes ago
Comment by bityard 1 hour ago
It is a very rare (and ballsy, and perhaps incompetent) company that ignores their lawyers' recommendations to adhere to the letter of all of the software licenses they are bound to.
Comment by ffsm8 1 hour ago
Comment by JumpCrisscross 1 hour ago
What was it?
Comment by drawnwren 1 hour ago
1 - https://x.com/Kimi_Moonshot/status/2035074972943831491?lang=...
Comment by Iolaum 1 hour ago
Comment by ericpauley 35 minutes ago
Comment by embedding-shape 1 hour ago
Comment by jhonof 1 hour ago
Comment by embedding-shape 1 hour ago
Comment by parodysbird 1 hour ago
Comment by embedding-shape 1 hour ago
Comment by eamag 1 hour ago
That's interesting!
Comment by whimsicalism 1 hour ago
Comment by rhdunn 19 minutes ago
f_gate(b,x) = b * tanh(x / b) * sigmoid(x)
f_up(b,x) = b * tanh(x / b)
Looking at the graph I wonder if this is to try and get the best of both GLU (better representation at higher values of x >~ 5) and SwiGLU (the value bump just before 0).Comment by whimsicalism 17 minutes ago
Comment by pinkmuffinere 1 hour ago
Comment by a-dub 1 hour ago
maybe some interesting theoretical work there around the rate of production of new knowledge itself and various mechanisms (human approaches, mechanistic approaches, etc).
Comment by storus 2 hours ago
Comment by whimsicalism 2 hours ago
Comment by eamag 1 hour ago
Comment by sosodev 34 minutes ago
If I understand correctly, it's distillation via having a teacher model score each of the student's tokens for a problem based on their own probabilities of generating that token at each step in the sequence. The reward/loss is then applied as RL.
The multi-teacher bit seems to imply they're distilling from multiple models. It's light on the details, but it seems like it could be part of distilling from frontier/closed models. Provided they calculate the logprobs, which OpenAI seems to allow via API but not Anthropic. Maybe they have a way of estimating the logprobs externally?
This method can be used to learn any domain from the teacher. Biology included.
Comment by mike_hearn 10 minutes ago
Comment by porridgeraisin 38 minutes ago
Comment by Tejush 26 minutes ago
Comment by weberer 1 hour ago
Comment by colesantiago 1 hour ago
Kudos to Moonshot for truly being what OpenAI should have been.
Fable-level and frontier AI should be open source and available to everyone for free.
Comment by embedding-shape 1 hour ago
Kudos to Moonshot for making these weights available for download. Lets not fool ourselves and claim these are "open source" by any understanding of the concept though, there are usage restrictions (even if you download them) and also training data isn't clearly broken down either, nor it it actually using a FOSS license.
Comment by lossolo 50 minutes ago
They will probably never release the training data because that represents a large part of their competitive moat. The same is true of US companies (Google, OpenAI, Meta etc) none of which has released the full training data for its open models.
They use private datasets that cost a lot to acquire, synthetic datasets and a lot copyrighted material for which they don't have licensing.
What matters most is that, with the necessary hardware, I can download a near frontier model, run it and modify it however I want. The other concerns you mentioned are just noise. And if my company is generating $20 million in revenue or serving 100 million users, it can probably afford a relatively inexpensive commercial license.
Comment by embedding-shape 36 minutes ago
Yeaaah, and this, of course, is worth it, because it leads to you being able to download a near frontier model. Don't get me wrong, long-term humanity is probably better of with science with little regards to pesky things like ethics and provenance, but we also have a tendency to not fully realize the downstream or wider effects until way too late.
And sure, there is a lot of reasons to go with keeping your software proprietary too, I'm not trying to claim otherwise, same with training data. It's just that usually we don't call proprietary software "open source" unless it is open source, regardless of the reasons someone keep it proprietary or not, could be for whatever reason really.
Comment by bilbo0s 37 minutes ago
Not that what you've written isn't the case. However, in addition to what you've written, (or probably even before what you've written), there's the fact that everything they're training on is stolen IP. Same with US LLM labs.
Let's not kid ourselve's about where the training data is coming from. They are not asking artists, writers, coders, content creators, etc etc etc for permission to use their creations.
Anthropic, Moonshot et al are doing incredible things, but we shouldn't gloss over the costs. Both present and future costs are kind of enormous.
Comment by dboreham 34 minutes ago
Comment by embedding-shape 24 minutes ago
- Proprietary - No access beyond remote endpoints
- Downloadable - You can run it, but there are restrictions and training data/code isn't public and/or under FOSS license, nor are the weights under a FOSS license
- Open weights - The weights are under a FOSS license and downloadable without restrictions, but not all of training code/data is public or under a FOSS license
- Open source - The model architecture, weights, training code and data are all under a FOSS license, and you can freely download and use them without any restrictions
Few models hit the mark for "open source model", many models people call "open weights" would go under "downloadable" here, which personally I think would be accurate.
Comment by lenerdenator 1 hour ago
Comment by embedding-shape 1 hour ago
Personally I was wishing/hoping for one of the recent Gemma releases to be in the ~100B class at least, but sadly Google is keeping that all for themselves.
Comment by Stagnant 51 minutes ago
Comment by layer8 1 hour ago
Comment by embedding-shape 1 hour ago
Comment by lossolo 58 minutes ago
Comment by whimsicalism 48 minutes ago
Comment by embedding-shape 34 minutes ago
Deeply fun "care about improving people's lives" quote on your user page :p
Comment by whimsicalism 21 minutes ago
Comment by segmondy 1 hour ago
Comment by boomskats 1 hour ago
Comment by chrsw 1 hour ago
Comment by brcmthrowaway 1 hour ago
Comment by abratabia 1 hour ago
Comment by histiq 1 hour ago
Comment by tudou527 1 hour ago
Comment by kiaansaraiya 1 hour ago
Comment by samxli 1 hour ago
Comment by m00dy 2 hours ago
End-to-end tokens/sec and cost on realistic coding agent trajectories, including tool outputs and retries, not isolated decode benchmarks.
Cache hit rates and prefill cost for branching, multi-turn sessions.
Router-load distributions after post-training, where expert collapse or specialization problems often show up.