A new “open” version of a language model is set to be launched this year, enabling developers to operate the model on their own systems.
In a recent update shared on X, the CEO mentioned that the artificial intelligence company plans to introduce a powerful “new open-weight language model with enhanced reasoning” in the upcoming months. However, they first intend to collect input on “how to ensure it is maximally beneficial.”
“This has been on our minds for quite some time, but other priorities took precedence. Now, it feels crucial to move forward,” he remarked, noting that this marks the first “open-weight” model since the release of GPT-2 in 2019.
He added, “We still have some choices to make, which is why we are organizing developer events to gather insights and subsequently experiment with early prototypes.”
Image Credit:
An open-weight language model will be accessible to the public for use, downloading, modifying, or deploying according to individual needs. While it is not as open as a fully open-source model, it represents a shift from GPT-3 and GPT-4, which were completely closed off.
The CEO announced that developer events for the model will commence in San Francisco, California—where the company is based—within the next few weeks, followed by sessions in Europe and the Asia-Pacific region.
Image Credit:
“We’re eager to see what developers create and how large corporations and governments choose to utilize it, especially when they prefer operating a model themselves,” he said.
“We will put in extra effort knowing that this model will likely be modified after its release.”
First “open” model since GPT-2 in 2019
The open-weight language model, GPT-2, was partially launched in February 2019, followed by its complete release in November of the same year.
Related: ‘Our GPUs are melting’ — A limiter was implemented after a surge in demand
The CEO also mentioned on February 12 that there are plans to roll out GPT-4.5 and GPT-5 in the upcoming weeks or months.
The competition in AI has intensified with the introduction of a rival model that operates similarly to ChatGPT but was reportedly developed at a significantly lower cost and within a shorter timeframe.
Additionally, a new open-source AI model designed for cost-efficient AI agents was released by a major tech company on March 26, while another prominent player revealed Gemini 2.5, its latest experimental AI model, on March 25.
Meanwhile, a notable tech CEO shared on March 19 that their AI model family, released in February 2023, has reached a milestone of 1 billion downloads.
Magazine: ‘Chernobyl’ illustrated the need to address AI risks, according to recent discussions