Key Takeaways
- OpenAI plans to release a new open-weight AI model.
- Open-weight models can be downloaded and deployed locally but aren’t fully open-source.
- OpenAI has not published detailed code and documentation for an AI model since GPT-2.
In the coming months, OpenAI plans to release what Sam Altman described as a “powerful” open-weight model—its first since GPT-2.
The move will see OpenAI partially return to its roots in open research. However, the company is stopping short of fully open-sourcing the new AI.
Open-Source vs. Open-Weight
In machine learning, model weights refer to the numerical parameters AI models learn during training.
Disclosing model weights means third parties will be able to run the model locally rather than relying on OpenAI’s servers.
In contrast, open-source projects publish model weights alongside the code and documentation that describes how the model works.
As well as being able to deploy models on their own hardware, the open-source approach means other parties can modify and adapt AI to suit their needs.
The Open-Source Rennaisance
The timing of OpenAI’s announcement on Monday, March 31, coincides with the recent explosion of open-source AI models like DeepSeek R1.
For the first time in recent years, open-source models offer comparable performance to the most advanced proprietary AI.
In China especially, DeepSeek’s success has revitalized the open-source AI scene, giving it new momentum after a period in which closed research dominated.
However, large American firms, including OpenAI, maintain strict control over their most powerful models.
How OpenAI Embraced Closed AI
When OpenAI was founded in 2015, what was then a non-profit looked remarkably different from the company today.
“Our aim is to build value for everyone rather than shareholders,” OpenAI said in its founding statement.
“Researchers will be strongly encouraged to publish their work, whether as papers, blog posts, or code, and our patents (if any) will be shared with the world,” OpenAI added.
What a difference a decade makes.
Since 2015, OpenAI has filed over 150 AI-related patents. Its latest research papers offer only a glimpse into the inner workings of its technology.
Meanwhile, former employees have spoken out against an increasingly restrictive research environment that limits what they can publish.
This transformation has inevitably drawn criticism.Most notably from co-founder Elon Musk.
Musk sued OpenAI in 2024, alleging that he and “the non-profit’s namesake objective was betrayed by [CEO Sam] Altman and his accomplices.”
In March 2025 the court rejected Musk’s request for a preliminary injunction, a decision that was welcomed by OpenAI.
James Morales is CCN’s blockchain and crypto policy reporter. He has been working in the news media since 2020, writing about topics such as payments, banking and financial technology. These days, he likes to explore the latest blockchain innovations and the evolving landscape of global crypto regulation.
With an educational background in social anthropology and media studies, James uses his platform as a journalist to explore how new technologies work, why they matter and how they might shape our future.
