- OpenAI will release a new open-weighted language model with advanced reasoning.
- It will allow local execution and some level of customization, although it will not be completely open source.
- Altman acknowledges a change in strategy at the company, under pressure from competitors like Meta and DeepSeek.
- OpenAI will host global events to gather community feedback prior to launch.
OpenAI, the company behind popular artificial intelligence developments like ChatGPT, has confirmed that it is working on the launch of a new language model that will mark a turning point important in its technological opening policy. This model, described as “open-weight”, will include advanced reasoning and development capabilities can be run on local devices, which represents an important step towards greater decentralization of the use of AI.
Sam Altman, executive director of OpenAI, explained that this movement It represents a significant change from the closed approach that the company has maintained in recent years.. Although it is not a completely open source model, it will allow developers to work directly with the trained weights of the neural network, facilitating its adaptation to different needs without exposing sensitive internal details such as data sets or the entire architecture.
A new, more accessible, but not completely open model
The open-weight model that OpenAI is preparing will be the first with these characteristics that the company has launched since GPT-2 in 2019Unlike that initial experiment, the new model promises advanced reasoning skills, comparable to those of the o3-mini model, also from OpenAI, consolidating itself as the highest-performance proposal within the universe of semi-open models.
In practical terms, it will offer an intermediate opening: Developers will be able to download the weights, fine-tune them, and integrate them into their own solutions., but without access to the original training data or the internal technical structure. This ensures control over its use, while allowing for a degree of flexibility and collaboration.
The decision comes amid growing competitive pressure. Models like Meta's Llama and Google's Gemma have gained traction in the ecosystem thanks to a more open approach. In fact, OpenAI acknowledges that for years it has maintained an overly closed stance, something Altman described as “being on the wrong side of history”.
Listening to the community: events to gather input
In order to adapt the new model to the real expectations of technical users, OpenAI has initiated a round of direct collaboration with the communityThrough a feedback form on its website, it's asking for feedback on what features you'd like to see implemented in the new tool.
In addition, the company plans to develop in-person events Testing and developer meet-ups. The first sessions will take place in San Francisco, and will later expand to cities in Europe and the Asia-Pacific region. These meetings will allow for initial prototypes to be showcased and suggestions for improvements to be collected before the official launch.
Steven Heidel, a member of the OpenAI technical team, confirmed that the model will be able to run on local hardware. This detail is relevant, since eliminates exclusive dependence on cloud services, giving way to solutions with greater autonomy and privacy.
Risks, control and ethical framework of use
Enthusiasm for more accessible models coexists with concerns about their misuse. Researchers and security experts have warned that such powerful tools, which can be modified after publication, could facilitate malicious use.
In response to this, members of the OpenAI security team, such as Johannes Heidecke, have stressed that the model will go through a rigorous vetting process. assessment according to the "Preparedness Framework" of the company. OpenAI assures that it will not release any version that poses a catastrophic risk to society.
Additionally, implicit limits could be placed on the license, restricting activities such as reverse engineering or use in sensitive contexts in certain countries. For now, OpenAI has not specified details about the license under which the model will be distributed.
A response to an increasingly competitive industry
The planned launch of the open-weight model comes at a time when the race for AI supremacy has intensified. DeepSeek, a Chinese AI lab, has garnered significant attention with its fully open R1 model, which is capable of matching or even surpassing some of OpenAI's proprietary models at a fraction of the usual development cost.
Likewise, Meta with its Llama family and Google with Gemma have consolidated proposals in the open-source field. This has forced OpenAI to rethink its strategy, opting for a partial opening so as not to lose traction against its main competitors.
Although technical details such as the number of model parameters, the context window or the training methods applied have not yet been revealed, Altman has assured that it will be a model “very competitive in reasoning”, which leaves the door open for it to position itself above other existing open alternatives.
What is an open-weight model really?
In the world of artificial intelligence, an “open-weight” model is one that makes his trained weights publicly availableThese weights are the numerical values generated during your training process and allow you to make predictions or generate text.
Unlike fully open models that also expose source code, training data, and architecture, open-weight models offer less transparency but greater usability than closed models accessible only through APIs.
This allows developers Customize existing models without starting from scratch, achieving specific adaptations at lower costs. It also facilitates local implementation, which is especially useful for companies or governments that require greater privacy.
Looking ahead: the model as a turning point
With this initiative, OpenAI not only seeks to compete technically, but also rebuild bridges with a developer community that values collaboration, openness and technological autonomy.
There is no official release date yet, but it is expected to happen sometime in the second half of 2025. In the meantime, Tests, events, and feedback will set the course. of what could be one of OpenAI's most ambitious bets since its inception.
While not completely open, the decision to release an open-weight model with advanced reasoning capabilities represents a turning point. important strategic for OpenAIIn an increasingly competitive environment with increasingly open models, this move seeks to position the company on a middle ground between absolute control and transparency, maintaining leadership without losing contact with the global technical community.
I am a technology enthusiast who has turned his "geek" interests into a profession. I have spent more than 10 years of my life using cutting-edge technology and tinkering with all kinds of programs out of pure curiosity. Now I have specialized in computer technology and video games. This is because for more than 5 years I have been writing for various websites on technology and video games, creating articles that seek to give you the information you need in a language that is understandable to everyone.
If you have any questions, my knowledge ranges from everything related to the Windows operating system as well as Android for mobile phones. And my commitment is to you, I am always willing to spend a few minutes and help you resolve any questions you may have in this internet world.