This comment would be much better without the second line
ike_a 53 minutes ago [-]
I'm not sure I understand the case for open-source models being decelerationist, is this it?
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?)
sosodev 16 minutes ago [-]
I think the argument is that decentralization leads to deceleration because it means less centralized funding and data. Those are the two primary ingredients for accel.
The problem with the decel/accel rhetoric is that it lacks nuance.
StevenWaterman 49 minutes ago [-]
I think it's basically open weights => more inference competition => less profit from inference => less training competition
f311a 27 minutes ago [-]
> less training competition
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.
Iolaum 49 minutes ago [-]
Open Source models decelerate growth of closed AI. For people who think (or want) AI = closed_AI then that argument has weight. Good luck getting them to update their priors.
zozbot234 40 minutes ago [-]
Open source AI is actually a lot less "powerful" than genuine frontier models, i.e. it has a much tighter inherent capability ceiling. This is "decelerationist" from a purely AGI-pilled point of view but it's actually great if you're worried about a capabilities arms race putting AI Safety at severe risk.
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?
ike_a 16 minutes ago [-]
Do you think they are inherently less powerful? I'd imagined that internal labs have a head start / more funding so the labs are playing catch-up.
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?
zozbot234 10 minutes ago [-]
If you're targeting widespread local/on prem deployment as many open weight models are doing, that inherently limits your scale in terms of total model weights/inference-time compute compared to running in a few centralized datacenters. A centralized model will always be able to leverage a larger scale of deployment, placing it much closer to the genuine "frontier".
viccis 14 minutes ago [-]
I know it's a hard ask on this site, but I need you to start parsing content and not tone. It was a helpful bit of context, even if it was a bit vitriolic.
Almondsetat 11 minutes ago [-]
Why should I waste time parsing content and not tone? Why can't the commenter just avoid the tone? It even saves time since you can write less!
> 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
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
zozbot234 24 minutes ago [-]
To be clear, the restriction on "Model as a Service" past $20M yearly revenue is new to K3. K2.x had the attribution requirement for any commercial use with more than $20M monthly revenue.
simonw 22 minutes ago [-]
Good catch, thanks.
lossolo 2 minutes ago [-]
In what way was the license "janky"? Moonshot isn't a trillion dollar US behemoth. If these terms allow it to release near frontier models with open weights, so startups and anyone with enough hardware can use them, while charging only companies with more than $20 million in revenue or 100 million users, that seems like a reasonable trade off.
throwaway27448 47 minutes ago [-]
I wonder how they'll figure out who to target for litigation when this license is violated.
bityard 6 minutes ago [-]
Software licenses aren't enforced through litigation as much as they are enforced through the _threat_ of litigation and legal risk. In other words, pretty much every company pays lawyers to minimize legal risk. Those lawyers inevitably look at all the contracts, agreements, and software licenses, and tell the C-suite what to do in order to keep their legal exposure as low as possible. "Don't violate other companies' IP," is pretty low-hanging fruit in those conversations.
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.
paxys 34 minutes ago [-]
How many companies host and serve models via API and have a $20M+ revenue? Going to be pretty straightforward to catch offenders.
ffsm8 43 minutes ago [-]
They haven't litigated the last public non-compliance... Despite that one being extremely public. So probably not at all for now.
JumpCrisscross 39 minutes ago [-]
> They haven't litigated the last public non-compliance
What was it?
drawnwren 36 minutes ago [-]
Cursor was thought to be but they were later found to be using an authorized provider [1]
Maybe because a non public agreement was in place?
46 minutes ago [-]
embedding-shape 26 minutes ago [-]
Hah, so much for "open weights" :D Fair enough, they're at least downloadable, shame they didn't end up being actually open, nor open source, the community had really high hopes for this. But again, still available for download, so better than nothing else I suppose.
jhonof 22 minutes ago [-]
Open source projects frequently have separate commercial licenses no? This isn't abnormal.
embedding-shape 19 minutes ago [-]
No, I don't know a single project I'd call "open source" (as understood by the FOSS community) that restricts what you can do with it, that'd make it very much "not open source" as you're discriminating against specific persons/groups/fields of endeavor.
parodysbird 11 minutes ago [-]
GPL licenses also restrict what you can do with it...
embedding-shape 7 minutes ago [-]
Beyond reciprocity which is the entire point of that license, what restrictions does it come with? I guess you could say that it has a restriction of adding new restrictions, but surely that's not what you're talking about?
GodelNumbering 5 minutes ago [-]
Back of the envelope calculation (could be off, correct me if I am)
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!
eamag 42 minutes ago [-]
> we build a self-evolving, hierarchically organized knowledge graph that agents continuously expand through web-scale exploration across knowledge-intensive and coding domains
That's interesting!
weberer 8 minutes ago [-]
Does anyone know if a torrent is available? I think it would take quite a while to download 1.5tb from their servers.
a-dub 46 minutes ago [-]
knowledge graph guided task synthesis. very cool! i have long wondered about the "how do you get good coverage of all the tasks" problem.
maybe some interesting theoretical work there around the rate of production of new knowledge itself and various mechanisms (human approaches, mechanistic approaches, etc).
whimsicalism 49 minutes ago [-]
It's funny that we've finally returned to tanh activation functions, time is a circle.
pinkmuffinere 1 minutes ago [-]
Wow this is fascinating enough that I’m actually going to read tfa lol
storus 1 hours ago [-]
What would be the current best method to fine-tune it for my own specific agentic tasks? LoRA + DPO? GRPO? Something else?
whimsicalism 1 hours ago [-]
LoRA + SFT, but it'll be big - better to wait for a finetuning API from one of the providers, I wouldn't jump straight to RL or off-policy pseudo-RL like DPO.
14 minutes ago [-]
eamag 44 minutes ago [-]
Can someone explain what are teachers in Multi-Teacher On-Policy Distillation? I can imagine math, coding and other verifiable domains, but they also have biology? Is it where distillation from bigger models come in?
lenerdenator 35 minutes ago [-]
What would it take to get an American open model to compete with this?
embedding-shape 23 minutes ago [-]
Latest "big" release from any of the bigger American lab must have been GPT-OSS-120b I think? Released ~summer 2025, so pretty much two years ago. Doesn't seem like it'll happen by itself, so something either forcing their hand figuratively, or something forcing their hand literally.
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.
layer8 16 minutes ago [-]
August 2025.
embedding-shape 14 minutes ago [-]
Bah, of course, thanks :)
segmondy 8 minutes ago [-]
if it's to be believed, it's so easy. you distill it, and you have a copy in 2 weeks, isn't that what China is doing? so 2 weeks from now, we should have one.
boomskats 28 minutes ago [-]
An act of G̶o̶d̶ Congress?
chrsw 25 minutes ago [-]
Something beyond my imagination
brcmthrowaway 20 minutes ago [-]
Is this p-hacking?
colesantiago 43 minutes ago [-]
This is amazing to witness.
Moonshot open sourcing Kimi K3, a frontier AI and other components really means we are getting abundant AI for all of humanity.
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.
embedding-shape 22 minutes ago [-]
> Kudos to Moonshot for truly being what OpenAI should have been.
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.
abratabia 9 minutes ago [-]
[flagged]
histiq 44 minutes ago [-]
[flagged]
tudou527 41 minutes ago [-]
[flagged]
kiaansaraiya 41 minutes ago [-]
[dead]
samxli 59 minutes ago [-]
[flagged]
m00dy 1 hours ago [-]
I would want to see three things before drawing strong conclusions:
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.
Rendered at 16:45:40 GMT+0000 (Coordinated Universal Time) with Vercel.
Anyone who claims open source and open weights models are "decel" needs to get their head checked
https://github.com/MoonshotAI/MoonEP
https://github.com/kvcache-ai/AgentEnv
https://github.com/MoonshotAI/FlashKDA
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?)
The problem with the decel/accel rhetoric is that it lacks nuance.
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.
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?
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?
So you're admiting you were trolling?
> 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
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
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.
What was it?
1 - https://x.com/Kimi_Moonshot/status/2035074972943831491?lang=...
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!
That's interesting!
maybe some interesting theoretical work there around the rate of production of new knowledge itself and various mechanisms (human approaches, mechanistic approaches, etc).
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.
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.
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.
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.