K-Culture GlossaryㅇWords from the people who build
Open Source
A way of releasing the code behind something so anyone can use, modify, and share it. In AI news, "open source release" often means something a bit different — worth watching for.
In plain words
Open source means publishing the underlying code of software — its blueprint — so that anyone can take it, modify it, and redistribute it. Think of it like a cooking analogy: instead of just selling the finished dish, you publish the entire recipe. Other chefs can use that recipe to run their own business, or tweak it and share their version. Most of the servers that run the internet, and the core of Android, were built this way.
But "open source" in AI news is often used to mean something a bit different, so it's worth being careful. Most articles about an "open source model release" aren't actually sharing the recipe (training code and data) — they're only sharing the finished dish (the model weights). Strictly speaking, this should be called "open weight," and there's ongoing debate in the industry over what really counts as open source.
Still, the power of releasing things openly is the same either way. Meta's Llama, DeepSeek, and Alibaba's Qwen have all drawn developers worldwide by releasing their models for free, becoming the biggest counterweight to companies selling paid models. This is also why companies give away things that cost a fortune to build — whoever captures the ecosystem first becomes the standard.
How it shows up in the news
"DeepSeek released its new model as open source" means anyone can download it and run it on their own servers. But usually what's released is just the weights — it's rare for the entire development process to be made public.
Try it yourself
- Go to github.com and search for the name of a service you like — you might be surprised how much code is fully public.
- Open any project and click through its files. Being able to see the full blueprint without even logging in — that's open source.
- Ask a chatbot: "Explain the difference between open source and open weight using a cooking analogy." This clears up one of the most confusing points when reading AI news.
Open Source vs. Open-Source Model: What's the Difference
Is the how-to-build process disclosed?, Is only the finished product disclosed?
This is one of the most commonly misused terms in AI news. Most articles that say something was 'released as open source' actually just released the finished model file, without releasing the data or training code used to build it. News outlets follow the terminology companies use, but it diverges from the original meaning.
| Aspect | Open Source | Open Weights |
|---|---|---|
| What's disclosed | The entire blueprint — code, process, conditions | The finished model file (weights) |
| Can it be reproduced? | Yes, you can take it and rebuild the exact same thing | You can use it, but you don't know how it was made |
| In AI news coverage | Rarely used in its original sense | Most things called an 'open-source model' actually belong here |
| Usage restrictions | Freely modify and redistribute | Often comes with usage terms, such as restrictions on commercial use |
| Analogy | Sharing the entire recipe | Handing over the finished dish |
Rule of thumbIf what's disclosed is 'how to build it,' it's open source; if it's 'the finished file,' it's open weight. It's worth knowing that news articles often don't distinguish between the two.
See also
Stories using this term
No story has used this term yet. New ones attach here automatically.