OpenAI Launches GPT-OSS: Free AI Models that Can Run On Your Laptop

Imagine a powerful AI model that can be downloaded for free, modified at will, and even run on an ordinary laptop. This isn't a fantasy; it's OpenAI's newly released GPT-OSS. Open, free, and controllable, a new era of AI is quietly dawning.

To adapt to different devices and usage scenarios, OpenAI divides GPT-OSS into two versions:

• The bigger version, with 120 billion parameters, can run on a high-performance Nvidia GPU, achieving performance close to that of OpenAI's o4-mini model, making it suitable for specialized tasks requiring high inference capabilities.
• The smaller version, with 20 billion parameters, is optimized for local deployment and requires only 16GB of RAM, running smoothly on a typical laptop. Its performance is comparable to that of the o3-mini.

Whether you're building enterprise applications or developing experimental projects on your own device, you'll find a version suitable for you. Furthermore, OpenAI has made both model versions available on major platforms like Hugging Face, Databricks, Azure, and AWS under the Apache 2.0 open source license, meaning you can freely modify, commercialize, and even embed them into your own products.

 

This isn't just the release of a new model; it also marks a significant shift in OpenAI's openness over the past six years.

It's important to note that since GPT-2, OpenAI hasn't released any open-weight models—GPT-3, GPT-4, and even the O series remain closed. This means that GPT-OSS is the first time since GPT's inception that anyone can publicly access the code that compromises the core of GPT-OSS.

OpenAI CEO Sam Altman made it clear in public until earlier this year that they would not be releasing open models due to security concerns. Nevertheless, reality has quickly evolved. With open source models like LLaMA, Mistral, Gemma and DeepSeek being released, the developer community has quickly shifted, and requests for local control, and ability to deploy in flexible ways, have substantially increased.

After DeepSeek attracted attention, Altman's stance shifted. He bluntly stated that OpenAI's refusal to release its own open models puts them "on the wrong side of history." This statement carries a weight of strategic reflection, signifying OpenAI's final realization that if it continues to cling to its closed-off approach, it risks being left behind by the true "open source wave."

The emergence of GPT-OSS is a response to this context. It both corrects past hesitations and reaffirms its open stance in the new competitive landscape.

This time, OpenAI isn't just "participating in openness"; it aims to re-establish its technological influence within the AI community with GPT-OSS.

Unlike traditional open-source models, GPT-OSS offers dual capabilities: it can be deployed independently in a local environment or integrated into existing service systems through OpenAI's APIs, enabling tasks such as reasoning, web browsing, programming, and agent control. It serves as both a powerful tool for independent development and a crucial interface for connecting to the OpenAI technology ecosystem.

OpenAI researcher Chris Cook admitted at a press conference, "Many people may not know that most of our customers have already been using various open models." This statement surfaces a reality: open source acceptance in the market has been slowly increasing, and OpenAI has been catching on to this "proven trend."

He added, "We hope to fill this gap and enable them to fully utilize our technology." This isn't just a product launch; it's an ecosystem restructuring—OpenAI aims to fully encourage users who already use open models to return to its own platform and ecosystem.

Of course, openness also carries risks. In response to long-standing security concerns, OpenAI has been particularly cautious in releasing GPT-OSS.

Officials stated that this is the most rigorously tested model to date, undergoing not only multiple rounds of internal review but also external security agencies to assess the model to ensure it is not misused in highly sensitive areas such as cybersecurity and bioweapons.

To make the model's behavior more transparent and controllable, OpenAI introduced a "Chain of Thought" mechanism in GPT-OSS. This means that when the model gives an answer, its thought process and reasoning path are displayed. This not only makes the results easier to understand, but also provides strong support for identifying potentially misleading or abnormal outputs.

Notably, GPT-OSS output is plain text, preventing "covert dissemination" in the form of images or code. Furthermore, as with previous OpenAI models, its training data is not publicly available, in part to protect the model's security perimeter and prevent reverse engineering and poisoning attacks.

This shows that even with open models, OpenAI is still struggling to draw the line between "safety and freedom."

OpenAI has not provided any benchmark data comparing GPT-OSS with other popular open models like Llama, DeepSeek, or Google's Gemma, so there is still a lot to be desired.

Nevertheless, OpenAI indicated that both versions of GPT-OSS performed as well as their does not disclose all reasoning model evaluations on tasks such as coding tasks and most awaited of the "Humanity's Last Exam," indicating that their performance is at level of Industry-leading standards. OpenAI co-founder Greg Brockman added that he was also very impressed, stating, "These models are great, and the team's work was impressive."

This has undoubtedly garnered GPT-OSS more attention and excitement within the open model community, adding new variables to the future competitive landscape.

OpenAI hasn't yet made a firm commitment regarding the release schedule for future versions of GPT-OSS. However, they hope that this model will be widely adopted by small developers and businesses who want greater control over how their data is used.

As co-founder Greg Brockman said, "We firmly believe that innovation will follow when the technical barrier to entry is lowered. If people are given the freedom to explore and adapt, they will create many unexpected and exciting results."

This not only demonstrates the true significance of open models, but also reveals OpenAI's confidence in the continued prosperity of the future technology ecosystem - the more open it is, the more likely it is to inspire more unexpected creativity.