OpenAI's code is mostly closed, despite the name

OpenAI is not open source in the way most people understand the term. The company's name is misleading — it refers to the original mission to develop artificial intelligence openly, not to the licensing of its software. Most of OpenAI's products, including ChatGPT and GPT-4, are proprietary. You cannot see the code, modify it, or redistribute it.

OpenAI does release some open source projects, but these are the exception rather than the rule. Whisper (their speech recognition model) and some research implementations are available under open source licenses. However, the core models that power ChatGPT and the API you pay to use are closed source. You access them through OpenAI's servers on their terms, not your own.

This matters because it affects what you can do with the software. With closed source, you are dependent on OpenAI's decisions about pricing, availability, content policies, and how long they keep your data. You have no right to audit the code or understand exactly how it works.

Key Takeaways

  • ChatGPT, GPT-4, and OpenAI's API are proprietary software, not open source, despite the company's name.
  • OpenAI does release some open source projects like Whisper, but these are separate from their main commercial products.
  • Closed source means you cannot see, modify, or run the code yourself — you use it only through OpenAI's servers under their terms.
  • If you need true open source AI models, alternatives like Meta's Llama or Mistral are available under open licenses.

What "open source" actually means in software licensing

Open source software comes with a license that lets you see the source code, modify it, and share it with others. Common open source licenses include the MIT License, GPL, and Apache 2.0. Each has different rules about what you can do, but they all share one core principle: the code is public and you have legal rights to use it.

This is different from proprietary software, where the company keeps the code secret and you only get permission to use the compiled program under their terms. You cannot see how it works, cannot fix bugs yourself, and cannot run it on your own servers without permission.

OpenAI's name suggests openness, but the company chose the proprietary model for its main products. This is a business decision, not a technical one. They could release ChatGPT's code under an open license, but they do not, because the model's value partly comes from keeping it proprietary.

Which OpenAI projects are actually open source

Whisper is OpenAI's speech-to-text model, released under the MIT License in September 2022. You can download it, run it on your own computer, modify it, and use it in your own projects. The code is on GitHub and the model weights are public.

DALL-E 2 is not open source — you access it only through OpenAI's API. However, OpenAI has released some research code related to image generation under open licenses.

GPT-2, an older language model, was released in stages with increasing openness. The full model and code are now public, though it is much smaller and less capable than GPT-3 or GPT-4. This was partly a research decision and partly a safety decision — OpenAI released it gradually to study how people used it.

ChatGPT and GPT-4 are not open source. You cannot download them or run them yourself. You can only use them through OpenAI's website or API, and you must follow their terms of service.

How closed source affects what you can do

When you use ChatGPT, you are sending your text to OpenAI's servers. OpenAI's privacy policy says they use conversations to improve their models, unless you turn off training on your account settings. Even with that setting off, OpenAI keeps your data for 30 days for safety and abuse monitoring.

You cannot run ChatGPT on your own server or computer. You cannot modify how it works. You cannot see the code that decides what it outputs. If OpenAI changes their pricing, removes features, or shuts down the service, you have no recourse except to stop using it.

With open source software, you own a copy. You can run it offline, modify it, audit it for security problems, and keep using it even if the original developer stops maintaining it. You have more control and more privacy, because your data stays on your machine.

Open source AI alternatives to OpenAI

Meta's Llama 2 is a large language model released under a free license. You can download it, run it on your own hardware, and use it for commercial purposes. It is smaller than GPT-4 and less capable, but it is genuinely open source.

Mistral is a French AI company that releases open source models. Their 7B model is small enough to run on a laptop and is available under the Apache 2.0 license.

Hugging Face hosts thousands of open source models contributed by researchers and companies. Many are language models, image models, or audio models. You can download them and use them freely, though you should check the specific license for each one.

These alternatives are not as polished as ChatGPT and may be less accurate for some tasks. But they give you privacy, control, and the ability to modify the code. The tradeoff is that you often need more technical knowledge to set them up and run them.

Why OpenAI chose the proprietary model

OpenAI's business model depends on charging for access to their models through an API or a subscription service like ChatGPT Plus. If they released the code and model weights as open source, anyone could run it for free on their own servers. OpenAI would lose their revenue.

There is also a safety argument. OpenAI says that keeping the model proprietary lets them control how it is used and prevent misuse. An open source model could be modified to remove safety guardrails or used for harmful purposes. This is a real concern, though it is also a reason some people distrust closed source AI — you cannot verify that the safety measures actually work.

The name "OpenAI" reflected the founders' original vision in 2015, when the company was a nonprofit focused on research. As OpenAI shifted to a for-profit model and built commercial products, the name became less accurate. The company kept it anyway, partly for brand recognition and partly because some of their research is still published openly.

How to check if software is actually open source

Look for the source code on GitHub or another public repository. If you cannot find it, the software is probably not open source. Check for a license file — it should be in the root directory and state clearly what you can and cannot do.

Common open source licenses include MIT (very permissive), Apache 2.0 (permissive with patent protections), and GPL (requires you to share modifications). If a company claims to be open source but does not have a recognized license, be skeptical.

For AI models specifically, check whether you can download the model weights (the actual data that makes the model work) or only the code. Some projects release the code but not the weights, which limits what you can do. True open source AI includes both.

Frequently Asked Questions

Can I use ChatGPT's output in my own project?

Yes, but with restrictions. OpenAI's terms say you own the output, but you cannot use it to build a competing service. You also cannot use it to train another AI model without permission. Check OpenAI's current terms of service for the exact rules, as they change.

Is GPT-2 safe to use if it is open source?

GPT-2 is smaller and less capable than newer models, so it is less likely to produce harmful content by accident. But like any open source software, you are responsible for how you use it. OpenAI released it partly to study how people would use an open language model.

Can I run Whisper offline on my computer?

Yes. Whisper is open source and you can download it from GitHub. You can run it locally without sending audio to OpenAI's servers. This makes it useful if you need privacy or do not want to pay per API call.

Why does OpenAI's name say "open" if they are not open source?

The company was founded in 2015 as a nonprofit with a mission to develop AI openly and share research. The name reflected that goal. As OpenAI became a for-profit company and built commercial products, the name stayed the same even though the business model changed.

If I want open source AI, what should I use instead?

Llama 2 from Meta and Mistral are both open source language models you can download and run yourself. Hugging Face hosts many others. They are less polished than ChatGPT but give you full control and privacy. You will need some technical knowledge to set them up.