Toward the end of March, OpenAI announced its plans to release its first “open” language model since the debut of GPT-2, aiming for a launch later this year. Details about this upcoming model are beginning to emerge from the company’s engagements with the AI developer community.
According to information from TechCrunch, Aidan Clark, the Vice President of Research at OpenAI, is spearheading the development of the new open model, which is currently in the very early stages. OpenAI is targeting an early summer release, with an aim to position the model as a benchmark leader among other open reasoning models, similar to OpenAI’s o-series models.
OpenAI is reportedly considering a highly permissive license for this model, with minimal usage or commercial restrictions, as conveyed by TechCrunch’s sources. Other open models, such as Llama and Google’s Gemma, have been criticized within the community for imposing demanding requirements, a criticism OpenAI appears keen to avoid.
OpenAI is under rising pressure from competitors, such as the Chinese AI lab DeepSeek, which have embraced an open approach to releasing models. Unlike OpenAI’s strategy, these rivals provide their models to the AI community for experimentation and, in some cases, commercialization.
For some companies, this strategy has proven extremely successful. Meta, for example, has heavily invested in its Llama family of open AI models and reported in early March that Llama achieved over 1 billion downloads. Likewise, DeepSeek has rapidly built a substantial global user base and caught the interest of domestic investors.
TechCrunch sources also indicate that OpenAI intends for its open model, which will operate on a “text in, text out” basis, to be compatible with high-end consumer hardware. It may also allow developers to toggle its “reasoning” capability, similar to models recently released by Anthropic and others, though reasoning could enhance accuracy at the expense of increased latency. If the initial release is successful, OpenAI may develop additional models, possibly including smaller versions.
In prior public remarks, OpenAI CEO Sam Altman acknowledged that the company may have been on the “wrong side of history” regarding open-sourcing its technologies. During a Reddit Q&A in January, Altman expressed the need to devise a new open-source strategy, although not everyone at OpenAI shares this view, and it is not the company’s highest priority at present. Nevertheless, they plan to produce superior models, albeit with potentially less of a lead than in previous years.
Altman also stated that OpenAI’s forthcoming open model would undergo thorough red-teaming and be assessed for safety. Sources informed TechCrunch that OpenAI plans to publish a model card for the model, outlining comprehensive technical details, internal, and external benchmarking, as well as safety testing results.
In a post on X last month, Altman asserted that before the model’s release, it would be evaluated under the company’s preparedness framework, similar to other models, with additional efforts due to anticipated post-release modifications.
OpenAI has faced criticism from some AI ethicists for allegedly rushing safety testing of recent models and failing to publish model cards for others. Altman has also been accused of misleading OpenAI executives about model safety reviews prior to his brief removal as CEO in November 2023.
Requests for comments from OpenAI have been made, and updates will be provided as they become available.
