New Openai models are not really open: What to know about open intelligence

Despite the name of the company, Openai has not dropped an open version of AI’s models since GPT-2 in 2019. This changed on Tuesday, as Sam Taman CEO. subscriber Two of the open weights, logical thinking models, called GPT -SS-120B (120 billion teachers) and GPT -SS-20B (20 billion teachers).

If the open weights are a new piece Ai Jargon To you, don’t worry. In the simplest possible terms, open weights are a category of artificial intelligence models that operate products such as Chatbots and generators and video. But they differ philosophically from the technology on which the tools of artificial intelligence are based now. Chatgpt, Gemini and Copilot are operated by closed models, which means that we do not have real visions on how these black boxes make. Open weight models give us a peek on the mechanical processor behind the curtain, if it is permissible to speak.

GPT -SS is a big deal. It is the most recent weights thinking model, with a strong performance in the real world similar to the O4-MINI, you can operate locally on your computer (or phone with a smaller size). We believe this is the best and most open model for …

Sam Al -Taman (Sama) Aug 5 2025

You do not need to be developed or an expert in machine learning to understand how these open models work or even to run them yourself. Here is everything you need to know about open weights and open source artificial intelligence models.

What is an open artificial intelligence model?

All artificial intelligence models have weights, which are properties or elements. Models are trained to give certain connections more weight or value.

The open weights model does exactly what its name-the weights is available to the public, as it is specified before Federal Trade Committee. The developers can see these weights and how to use them to create artificial intelligence models.

“It can be said, that the most valuable things in large models (language) are actually weights. You can do a lot if you have weights, which are somewhat different from traditional programs,” Omar KhatabAssistant Professor of Computer Science at the Massachusetts Institute of Technology and Researcher at the Computer Science and Artificial Intelligence Laboratory (CSAIL), Tell CNET.

For example, Chatbot is truly designed to predict the next logical word in a sentence. He was trained to link words in his outputs that appear repeatedly next to each other in their training data, and is supposed to be in a logical order. Words that appear next to each other can be given a greater weight than words that do not often appear next to each other.

These weights are just numbers but open weight models also come with a map.

“In open weights (models), you get the weights, which are the numbers, and you get how these weights are set in the structure of the nerve network, so the nerve network layers, in order to be able to start them already.” The structure of the model shows how the company commits its models, which are “incredible”.

Open weight models are aimed primarily with developers, who can integrate the model into current projects, such as helping to build artificial intelligence agents. For “committed amateurs”, as Khattab said, you can use the specifications to run the model locally on your laptop, which can help reduce the privacy concerns that come, for example, using AI through the company’s mobile phone application. Researchers will also have a clearer look on how artificial intelligence works internally.

How do you accumulate the new open Openai models?

The new open weight models are two sizes, 120 billion teachers (128 experts, a 128K window) and 20 billion teachers (32 experts but the same window of 128K). Experts refer to the number of sub -networks enjoyed by the model, and context windows describe the amount of information that the model can process and include in its responses. The largest figures of both indicate that a model is able to have more sophisticated answers and has a greater fire strength.

In terms of performance, Openai informed that the 120B model “achieved almost near” with The latest thinking model, O4-MINI On basic thinking standards while running on the 80 GB of graphics processing unit. The 20b-Moights Form has been performed similarly to the O3-MINI and run on the 16 GB device-and this means, this smaller model can be operated on well open weights on laptops and some smartphones. (Like all artificial intelligence models, it is operated locally, your speed will depend on the power of fire in your device.)

The models will be available under APache 2.0 license, a type of friendly license for open sources. You can check more in -depth specifications in The form card and A paper on safety trainingGet advice from Openai’s Development instructions Watch the weights for yourself now Lugingface and Jaytab.

Are open weights the same open source artificial intelligence?

Open weight models are associated with an open source AI, but they are not the same. The open source as a concept refers to the programs that do not have owners of ownership, the source code is generally available and can be used by most of any person under open source licenses. the Open source initiativeA non -profit institution calling for open source programs, identification AI is open source “a system available under conditions that gives users the freedom to use, study, modify and share (IT).”

Open-Beights AI model is not the same as an open source model. One way to think about the difference between the two with bread, Soba VasodivanThe CEO of Operations at Mozilla.org and Vice President at Mozilla Corporation told CNET.

“Open weights model is a person who gives you this baked cake and says:” Oh, it is made of flour, sugar and eggs. “These are the weights.

For open weight models, the types of things that are not revealed are the data on which the model has been trained and the symbol used to train it. Training data is a point of disagreement between artificial and human intelligence companies Creating content; Artificial intelligence companies are high -quality content and a human being to improve their models and improve them. Some companies collect these data through licensing agreements, but some publishers and creators have filed lawsuits that claim to be artificial intelligence companies Copyright content. (Disclosure: Zif Davis, the parent company CNET, filed a lawsuit against Openai, claimed that it had violated the copyright of ZifF Davis in training and operating artificial intelligence systems.)

Training data, regardless of its origin, is one of the most valuable things that the artificial intelligence company has. But it is likely that it is not included in any version of an open model.

“I think that because of the huge domain (data), because of responsibility, because many of them are licensed and not supposed to share it, I think it is fair that we assume that no companies will start at least a profit at least, at any time soon, perhaps at all.”

Vasodifan said that artificial intelligence is really open source with information available to the public. It may be difficult to re -train or inspect open weight models without additional information. “You are still a little bit in the dark about how to build it or what data you have formed,” Vasodifan said.

Why should I care about Amnesty International Open Source?

These differences between open and open source and closed models are likely to not affect your intermediate experience using Chatbot Amnesty International. But they are important qualities for developers who must be taken into account when choosing and using AI models, which are more broad for us to understand technology that infiltrates our lives online.

There is no guarantee that the internal works of open weight models will reflect what is inside their closed models, ruin said. But for the company whose product is synonymous with GEN AI, it is sure to have an impact on how he works on the people who use it and those who design it behind the scenes. When people start entering the weeds with new models, we will learn more and see the impact on the industry.

In general, the basic philosophy of the open source is that technology improves when more people can reach it, such as academics who can study it and security services who can detect weaknesses. Vasodifan said: “You have to let people build, expand and innovate their scope and be able to place holes and make them better.”

For more, check our Beginner guide to Chatgpt and How does Amnesty International change our research experience.