Altman Officially Announces Open Source: After GPT-2, We Will Release the First Open-Weight Model with Reasoning Capability

OpenAI is finally going to be “Open” again!

This morning,  Altman tweeted that OpenAI will release a powerful new open-weight model with reasoning capabilities in the coming months.

OpenAI hopes to communicate and collaborate with developers, researchers, and the community to explore how to maximize the potential of this model. OpenAI is very much looking forward to making it an outstanding model.

Open weights mean that the training parameters or weights of a language model are publicly accessible. Developers can use them to analyze and fine-tune the model to perform specific tasks without requiring the original training data.

Altman further stated that OpenAI is planning to release the first open-weight language model since GPT-2. OpenAI has considered this for a long time, but other priorities have delayed the process. Now they think it is very important to do this.

Before release, OpenAI will evaluate this model according to its Preparedness Framework, just as it does for any other model. And considering that the model may be modified after release, OpenAI will do additional work.

OpenAI still has some decisions to make, so it will hold a developer event to collect feedback and then allow developers to try out the early prototype. The event will start in San Francisco in a few weeks and then be held in Europe and the Asia-Pacific region.

The following are some of the issues that OpenAI hopes users will provide feedback on:

In addition, Steven Heidel, a researcher at OpenAI, said that this open-weight model will be able to run on users’ own hardware.

Johannes Heidecke, a member of the OpenAI model safety team, said, “Safety has been the core focus of the development of this open-weight model from pre-training to release. Although opening up the model presents unique challenges, we will be guided by a prepared framework and will not release models that we believe pose catastrophic risks.”

After the news that  Altman announced the open-source model spread, the comment section exploded. Some speculated whether o1-mini was going to be open-sourced, some wondered whether this model would be released before or after GPT-5, and others began to guess the name of the model.

And just one hour before the official announcement of the open-source model by  Altman, he excitedly declared, “Twenty-six months ago, the launch of ChatGPT was one of the craziest viral moments I’ve ever witnessed. We gained 1 million users in just five days. Now, we’ve added 1 million users in just the past hour.” This is certainly thanks to the explosive popularity of GPT-4o’s image generation.