In a major shift from its previous strategy, OpenAI has announced the launch of two new artificial intelligence models available for free and fully customizable. This move places the company in direct competition with Meta and China’s DeepSeek in the open-model arena, while sparking fresh debates over safety and openness.
Article contents:
- Launch of two “open-weight” models
- A shift in OpenAI’s strategy
- Echoing Zuckerberg’s words: “democratic values” and “broad benefit”
- “Open weight” vs. “open source”: What is the difference?
- Safety concerns
- Speculation surrounding GPT-5
- Conclusion: OpenAI embraces openness… on its own terms
- Frequently asked questions
Launch of two “open-weight” models
OpenAI, the developer of ChatGPT, has announced the release of two large language models with “open weights” that are free to download and allow developers to customize them. The models, named gpt-oss-120b and gpt-oss-20b-two, outperform similarly sized models in reasoning tasks, according to the company. This move represents a departure from the ChatGPT model, which relies on a “closed” system that cannot be customized.
A shift in OpenAI’s strategy
Despite its name, the majority of OpenAI’s models are closed systems. The company had not released an open model since 2019 when it introduced GPT-2. However, after China’s DeepSeek gained global attention with its open software, CEO Sam Altman acknowledged that the company needed to “find a different strategy for open source.” This announcement appears to be the first step in that new strategy.
Echoing Zuckerberg’s words: “democratic values” and “broad benefit”
Sam Altman stated that the company is excited to contribute to a growing collection of freely available AI models “grounded in democratic values… and for broad benefit.” This statement strikingly mirrors comments made by Mark Zuckerberg, who said that making models freely available ensures “that more people around the world can access the benefits and opportunities of AI” and prevents power from concentrating in the hands of a few companies. This demonstrates that the idea of “democratizing AI” has become a major battlefield among tech giants.
“Open weight” vs. “open source”: What is the difference?
OpenAI used the term “open weight,” which is a lower tier than “open source.” “Open weight” means that the model’s weights (the numerical parameters learned during training) are available, allowing developers to fine-tune and customize it. However, the training data and model architecture are not fully transparent. In contrast, true “open source” requires the disclosure of all these components.
Safety concerns
The existence of powerful, freely available AI models has raised concerns among some experts who warn that they could be adapted to cause severe harm, such as developing biological weapons. OpenAI stated that during testing, it created “maliciously fine-tuned” versions of the models to simulate biological and cybersecurity threats, but found they were “unable to access high-capability levels.” Nevertheless, debate over the safety of open models continues.
Speculation surrounding GPT-5
OpenAI’s announcement comes amid speculation that a new version of the model powering ChatGPT, namely GPT-5, could be released soon. On Sunday, Altman shared a screenshot of what appears to be the company’s latest artificial intelligence model, fueling excitement within the AI community.
Conclusion: OpenAI embraces openness… on its own terms
The release of these open-weight models represents a significant strategic step for OpenAI. It is an attempt to keep pace with market trends toward openness, compete with Meta, and satisfy developer calls for greater flexibility. However, by choosing the term “open weight” instead of “open source,” the company also signals that it still wishes to retain a degree of control over its core technology.
Frequently asked questions
Are these models as powerful as ChatGPT?
The larger model, 120b, achieves performance roughly on par with OpenAI’s o4-mini model in basic reasoning. They are powerful models, but they are not at the same level as the company’s latest and largest closed models, such as GPT-4o.
What can developers do with these models?
They can download them, customize (fine-tune) them on their own data, and run them on their own hardware. This allows for the creation of specialized AI applications tailored specifically to particular use cases.
Why is this move important?
Because it represents a shift in OpenAI’s strategy—a company that started as an open non-profit organization before becoming more closed. Its return to releasing open models (even partially) will increase competition and innovation in the field.
