from China. China doesn‘t seem very keen on open source as a whole to be honest. That is unless they can monetize on open source projects from outside of China. Their companies love doing that.
DeepSeek being an LLM is far from open source and especially not „truly“ open. The very article you linked basically says as much but wraps it in pretty words. Talking about ignorance.
Yes I found out I was wrong, and I thought I had edited most of the wrong posts claiming deepseek is open source.
You are right it isn’t, despite articles claiming it is.
The dataset is massive and impractical to share, and a dataset may include bias and conditions for use, and the dataset is a completely separate thing from the code. You would always want to use a dataset that fit your needs. From known sources. It’s easy to collect data. Programming a good AI algorithm not so much.
Saying a model isn’t open source because collected data isn’t included is like saying a music player isn’t open source, because it doesn’t include any music.
EDIT!!!
TheGrandNagus is however right about the source code missing, investigating further, the actual source code is not available. and the point about OSI (Open Source Initiative) is valid, because OSI originally coined the term and defined the meaning of Open Source, so their description is per definition the only correct one.
Open source as a term emerged in the late 1990s by a group of people in the free software movement who were critical of the political agenda and moral philosophy implied in the term “free software” and sought to reframe the discourse to reflect a more commercially minded position.[14] In addition, the ambiguity of the term “free software” was seen as discouraging business adoption.[15][16] However, the ambiguity of the word “free” exists primarily in English as it can refer to cost. The group included Christine Peterson, Todd Anderson, Larry Augustin, Jon Hall, Sam Ockman, Michael Tiemann and Eric S. Raymond. Peterson suggested “open source” at a meeting[17] held at Palo Alto, California, in reaction to Netscape’s announcement in January 1998 of a source code release for Navigator.[18] Linus Torvalds gave his support the following day
Still debatable, the weights are the code. That’s a bit like saying “X software is not open source because it has equations but it doesn’t include the proofs that they’re derived from”.
And LLM is simply such a bad example for Open Source in general. They couldn‘t have chosen a worse example to make their point. That‘s what’s frustrates me.
The tide has turned. With the December 2024 launch of DeepSeek’s free-for-all V3 large language model (LLM) and the January 2025 release of DeepSeek’s R1 (the AI reasoning model that rivals the capabilities of OpenAI’s O1), the open-source movement started by Chinese firms has sent shockwaves through Silicon Valley and Wall Street.
And:
DeepSeek, adopting an open-source approach was an effective strategy for catching up, as it allowed them to use contributions from a broader community of developers.”
I’ve read similar descriptions in other articles, seems your claim is false.
EDIT PS:
Turns out on further investigation that Deepseek is NOT open source, there is NO access to the source code for Deepseek. Only the weights as others have rightfully claimed.
Investigating further I can see it is NOT open source. All the articles saying that are lying, probably unknowingly just as I believed the claim, they probably did too, and I’m NOT being sarcastic!
I have no idea why publishing these “weights” is considered open source, it has nothing to do with Open Source as defined by OSI, which I believe has a historical right to the term.
Alibaba has released Qwen models under Apache licenses (and they are some of the best models that can reasonably be ran locally). Some argue that models aren’t really open source unless the training code and datasets are made available though.
They are releasing lots of open weight models. If you want to run AI stuff on your own hardware, Chinese models are generally the best.
They also don’t care about copyright law/licensing, so going forward they will be training their models on more material than Western companies are legally able to.
I don‘t know of a single
from China. China doesn‘t seem very keen on open source as a whole to be honest. That is unless they can monetize on open source projects from outside of China. Their companies love doing that.
Your ignorance is not a valid point.
https://techwireasia.com/2025/07/china-open-source-ai-models-global-rankings/
deleted by creator
DeepSeek being an LLM is far from open source and especially not „truly“ open. The very article you linked basically says as much but wraps it in pretty words. Talking about ignorance.
Yes I found out I was wrong, and I thought I had edited most of the wrong posts claiming deepseek is open source.
You are right it isn’t, despite articles claiming it is.
DeepSeek the software is open source.
same with qwen, ernie, minimax, and kimi
Unless the dataset, weighting, and every aspect is open source, it’s not truly open source, as the OSI defines it.
The dataset is massive and impractical to share, and a dataset may include bias and conditions for use, and the dataset is a completely separate thing from the code. You would always want to use a dataset that fit your needs. From known sources. It’s easy to collect data. Programming a good AI algorithm not so much.
Saying a model isn’t open source because collected data isn’t included is like saying a music player isn’t open source, because it doesn’t include any music.
EDIT!!!
TheGrandNagus is however right about the source code missing, investigating further, the actual source code is not available. and the point about OSI (Open Source Initiative) is valid, because OSI originally coined the term and defined the meaning of Open Source, so their description is per definition the only correct one.
https://en.wikipedia.org/wiki/Open_source
no,
your changing the definition of open source software. which has been around a lot longer than AI has.
source code is what defines open source.
what deepseek has is open weights. they publish the results of their learning only. not the source that produced it.
Still debatable, the weights are the code. That’s a bit like saying “X software is not open source because it has equations but it doesn’t include the proofs that they’re derived from”.
In the same way as an Excel spreadsheet containing a crosstab of analytics results is “the code.”
It’s processed input for a visualization/playback mechanism, not source code.
what has been published by deep seek is the music, not the software, the music.
And LLM is simply such a bad example for Open Source in general. They couldn‘t have chosen a worse example to make their point. That‘s what’s frustrates me.
https://techwireasia.com/2025/07/china-open-source-ai-models-global-rankings/
And:
I’ve read similar descriptions in other articles, seems your claim is false.
EDIT PS:
Turns out on further investigation that Deepseek is NOT open source, there is NO access to the source code for Deepseek. Only the weights as others have rightfully claimed.
can you show me the actual source code?
the human readable code, not the weights.
The (claimed) source:
https://github.com/deepseek-ai
Investigating further I can see it is NOT open source. All the articles saying that are lying, probably unknowingly just as I believed the claim, they probably did too, and I’m NOT being sarcastic!
I have no idea why publishing these “weights” is considered open source, it has nothing to do with Open Source as defined by OSI, which I believe has a historical right to the term.
I apologize.
forgive my snarkiness. its usually the quickest way to this realization.
It‘s open weights but definitely not
Feel free to blame the technology as a whole but open source doesn‘t make exceptions for AI models.
every major chinese model is open source
Alibaba has released Qwen models under Apache licenses (and they are some of the best models that can reasonably be ran locally). Some argue that models aren’t really open source unless the training code and datasets are made available though.
They are releasing lots of open weight models. If you want to run AI stuff on your own hardware, Chinese models are generally the best.
They also don’t care about copyright law/licensing, so going forward they will be training their models on more material than Western companies are legally able to.