- Openai published GPT-OS-20B and GPT-OS-15B as open models
- The GPT-OS-20B model can operate locally on a consumer PC
- These models allow users to see each stage of the AI reflection process, which makes them more transparent in their operation
OPENAI has just dropped two new models of AI, GPT-OSS-120B and GPT-OS-20B. Not only are they new, but these are the first open weight models from the Chatgpt creator from GPT – 2.
The smallest of the two – GPT -ASS -20B – is particularly remarkable to be light enough to operate on a decently specified consumer PC. If you have about 16 GB of RAM and a certain patience, you can load it, ask questions and see how it happens to answers. The larger 120b model always requires serious material or cloud support.
The two models are part of the new Openai push to encourage developers to play with models and even market them for the average user. For the first time in years, curious developers and individuals can download and execute OpenAi models on their own machines, inspect how they think and rely freely on them. They are available via hugs and AWS under the Apache 2.0 license.
Being open means that models provide a level of transparency and independence that most people have not had since Chatgpt has become viral for the first time. Real -time reasoning is visible throughout, and you can see how the “logic” of the model leads to its final options on how to react and how it makes this decision.
It is a big change for Openai. The company has spent several years restricting access to its most powerful tools, offering only API ending points and paid levels. Now, he returns a bit to the GPT-2 era, but with much more competent models. Even so, the lighter model is not something that everyone will rush to replace the Chatgpt application.
GPT-Ass came out! We have made an open model that works at O4-Mini level and works on a high-end laptop (WTF !!) (and a smaller on a phone). Super proud of the team; Large triumph of technology.August 5, 2025
Open weight access
The flexibility provided by new models could be a boon for Openai because the open approach becomes more popular. Deepseek, Meta and Mistral have all recently published open models. But most of them proved to be semi-open, which means that they are trained on unknown data or have limited terms and limits of use.
GPT-AS models are simple in the supply of weights and license, although training data remains owners. And the OPENAI GPT-AS models provide compatibility with the widely used interface of Openai, as well as a larger window on how the model takes the decisions that stand out.
Thus, when Deepseek models tend to underline the raw power and the relatively low performance of their models, Openai is more interested in explaining how the models work. This will sting the interest of many developers who try to learn or experiment. You can literally take a break and look at what is going on inside the model.
It is also a signal for the developer community: Openai invites people to build with its technology in a way that has not been possible for years. These models are not content to take out cat supplements; They offer a foundation. With the Apache license, you can refine, distill, integrate or wrap them in entirely new products. And you don’t need to send data to a third party to do so. Those who hope for a more decentralized AI ecosystem are likely to be satisfied.
For those of us who are less technical, the main point is that you might see AI applications that do very well without needing a subscription price on your part, or personalized tools that do not send your cloud data. With this version, Openai can claim to be willing to share more with the developers, if not quite everything.
For a company that has spent a large part of the last two years building closed systems and premium subscriptions, the publication of a pair of models also capable of open licenses resembles a major philosophical change, or at least the desire to make such a change which seems real.
And although these two models do not represent the thrust in the era of GPT-5, the chief of Openai, Sam Altman, has teased that even more announcements are on the horizon, it is therefore likely that the new generation model is still being prepared.
