Information on OpenAI's First 'Open Source' Model Revealed

In March, OpenAI said it planned to release its first open source language model since GPT-2 at some point this year. According to TechCrunch yesterday, citing sources familiar with the matter, more details of the open-source model have been revealed: OpenAI VP of Research Aidan Clark is currently leading the development of the model, and the project is in the very early stages of development; OpenAI is aiming to release the model in early summer this year; the open-source model will be similar to OpenAI's o-series of inference models, and OpenAI hopes that the model will outperform other open-source inference models in benchmarking. OpenAI hopes that the model will outperform other open-source inference models in benchmarking, and is exploring a "highly permissive" license for the open-source model, which is intended to be commercially free. openAI doesn't want to tarnish the model's reputation in the community by making it subject to burdensome requirements. The report notes that Meta Llama and Google's Gemma are models that, despite being open source, have been criticized by users of AI forums for their onerous requirements. The open-source model uses a "text input/output" interaction model, and will allow developers to switch the "inference" function on and off freely, with developers choosing the speed at which results are generated. It's worth noting that if the model is well received, OpenAI may also release a smaller model. Currently, OpenAI is under pressure from a number of competitors because of its open modeling strategy: DeepSeek, which is new this year and has exploded in popularity, and Meta Llama, which was downloaded more than 1 billion times in March, but all of these models share the common feature of being "open source," which has brought huge traffic and investor attention to the models mentioned above.