Saturday, March 14, 2026

Openai releases a free GPT model that can work on a laptop

Share

OPENAI releases a modern model of an open weight called GPT-OSS It can be downloaded for free, be adapted and even laptop on a laptop.

The model comes in two variants: versions worth 120 billion parameters and 20 billion parameters. A larger version can work on one NVIDIA GPU processor and occurs similarly to the existing O4-Mini OpenAI model, while the smaller version works like O3-Mini and only works on 16 GB of memory. Both versions of models are issued today via platforms, such as face hugging, Databicks, Azure and AWS as part of the Apache 2.0 license, which allows for widely modified for commercial purposes.

This is the first OPENAI model in over six years, years before the debut of Chatgpt. Until the beginning of this year, the general director of Altman himself gave the concerns about security as the main reason why he did not give subsequent. Meanwhile, developers came to open models due to their lower costs and the ability to configure. In January, after the creation of Deepseek, Altman He said it Opeli “was on the wrong side of history” without releasing its own open models.

Now OpenAI is mentioned with an open model, which, he says, can perform the tasks of reasoning, browse the network, writing code and operate agents via existing API interfaces of the company. “I think that many people are surprised, knowing that the vast majority of our clients already use many open models,” said Chris Cook, Openai researcher during the media briefing. “We wanted to connect this gap and let them use our technology all over the board.”

On the security front, OPENAI claims that GPT-OSS is the most stringently tested model and that it has been tested with external security companies to make sure that it does not pose a risk in areas such as cyber security and biological weapons. Mental chain or noticeable process used to obtain an answer, “monitoring incorrect behavior, fraud and improper use” was shown, in accordance with the company’s press release. His output is only the text and, like all OpenAI models, GPT-OSS training data is undisclosed.

“The band really cooked with it.”

Opeli did not share reference points comparing GPT-OSS with other open models such as Llam, Deepseek or Google Gemma. Both GPT-OSS variants work similarly to the closed models of OPENAI reasoning on coding and tests, such as the last examination of humanity. “These are amazing models,” said Greg Brockman, Openai co-founder. “The band really cooked with it.”

Opeli does not undertake to the schedule of issuing future GPT-OSS versions, but he hopes that the model will be used by smaller programmers and companies that want more control over the employ of their data. “We have always believed that if you lower the barrier in access, the innovation would simply increase,” said Brockman. “You let people break in and then do things that are extremely surprising.”

Latest Posts

More News