Developers now have the chance to provide input on OpenAI’s latest initiative. On March 31, the AI company opened applications for feedback sessions on an upcoming open-weight language model—only the second such model since OpenAI transitioned its LLMs to closed access after GPT-2. According to OpenAI CEO Sam Altman in a post on X, the model is set to launch “in the coming months.”
Open-weight models allow users to modify them. In generative AI, “weights” refer to the connections between digital neurons, and adjusting them can alter how the model associates and prioritizes different concepts.
Altman described the new model as a “reasoning” model, similar to OpenAI o1.
“We’ve been considering this for a long time, but other priorities took precedence. Now, it feels important to move forward,” Altman stated.
Developers Can Share Their Input
Developers in San Francisco will have the opportunity to attend an OpenAI event in the coming weeks to provide feedback and test early prototypes. Similar events are planned for Europe and the Asia-Pacific region, though Altman did not specify exact locations.
“Before release, we will evaluate this model using our preparedness framework, as we do with all our models. Given that this model will be modifiable post-release, we will conduct additional assessments,” Altman wrote.
OpenAI’s Feedback Form
OpenAI is seeking input from developers, researchers, and the broader community. Its application form requests links to GitHub and professional profiles, as well as details on previous experience with open models, desired features in an OpenAI open-weight model, and potential use cases.
“We’re eager to collaborate with developers, researchers, and the wider community to ensure this model is as useful as possible,” OpenAI stated.
Open Models in the AI Landscape
The definition of open-source AI remains somewhat ambiguous. The Open Source Initiative defines open-source AI as models that can be freely used, modified, studied, and shared.
As TechCrunch noted, OpenAI’s competitors have already made strides with open models. Meta allows developers to access Llama model weights, while DeepSeek has released its R1 and V3 models, along with portions of their code.
OpenAI may be betting that allowing developers to modify and run its model independently will provide a competitive advantage in the rapidly evolving AI industry.