OpenAI plans to release its first open-weight language model with reasoning capabilities since GPT-2 in the coming months.
According to Reuters, CEO Sam Altman said this earlier this week. The new model will enable developers to analyze and refine the trained parameters or so-called weights for specific tasks, all without the need for the original training data.
In contrast to open-source models, where the full source code, training data and methodologies are made available, an open-weight model only offers access to the aforementioned weights. OpenAI wants to talk to developers to determine how such a model can be used most effectively.
The first event on this topic will take place in San Francisco in a few weeks. It will be followed by meetings in Europe and the Asia-Pacific region. Because decisions still need to be made, the company is organizing these developer events to gather feedback and experiment with early prototypes later.
New OpenAI roadmap
In February, Altman indicated that OpenAI wants to simplify its AI products. The company will draw up a new roadmap for the latest models. In addition, OpenAI, which receives financial support from Microsoft, must transition to a commercial company by the end of the year to secure the full funding of 40 billion dollars. SoftBank Group is supervising this investment.
The start-up emphasizes that this transition is necessary to attract sufficient capital for the development of advanced AI models. In October last year, OpenAI already completed a financing round of 6.6 billion dollars.
OpenAI is facing increasingly fierce competition. This week, for example, Elon Musk announced that he is merging his media platform X and AI startup xAI. This will probably make it easier to attract capital on the market. In addition, Musk has stated that he is opposed to OpenAI’s transition to a (more) commercial company.