Open is releasing a new openweight model dubbed GPTOS that can be downloaded for free, customized, and even run on a laptop.
The model comes in two variations: 120 billion parameters and 20 billion parameter version. The bigger version can run on the same NVIDIA GPU and perform the same as the current Open O4-MINI model, while the smaller version demonstrates similarly like O3-Mini and only runs on 16GB of memory. Both models versions are being released today by a platform such as the Apache 2.0 License faces, data BRICS, Ezor, and AWS, which can help them extensively modify for commercial purposes.
Before the start of the Chat GPT, this is the Openi’s first openweight model in six years. By the beginning of this year, CEO Sam Ultman blamed security concerns as the main reason for not issuing follow -up. Meanwhile, developers have come to open models due to their low cost and customization. In January, after the height of Deep Sak, Altman said that the opening was “wrong in the wrong side of history” by not releasing his open model.
Now, Openi is re -serving himself with an open -weight model, which says he can perform the reasoning work, browse the web, write code, and run agents through the company’s current APIS. “I think a lot of people are really surprised to know that the majority of our users are already using a lot of open models,” said an open researcher, an open researcher, during a media briefing. “We wanted to plug this gap and allow them to use our technology on the board.”
On the Safety Front, the Open says GPT-OS is its most strictly experienced model to date, and that it was tested with external security firms to ensure that it does not pose a threat to areas like cybercularity and biological weapons. According to a press release from the company, the model chain, or the process used to reach a response, is used, “to monitor the model’s misconduct, fraud and misuse”. Its output is only text and, like all Openi models, is unknown to GPTOS training data.
“The team really cooked with it.”
The Open AI has not shared a benchmark to compare the GPTOS comparison with other open models such as Lama, Dipic, or Google’s Jema. Both types of GPT-OS perform like the open reasoning models of Openi, which are related to tests such as coding tasks and the final examination of humanity. “These are incredible models,” said Open Coofer Greg Brockman. “The team really cooked with it.”
Openi GPTOS is not pledging a release schedule for the future version, but it is expected that the model will use small developers and companies that want more control over how their data is used. Brookman said, “We have always believed that innovation only increases if you reduce access to access.” “You allow people to hack, then they will do things that are amazing.”
