OpenAI prepares to launch its first open-source reasoning AI model
text_fields
OpenAI is gearing up to release its first open-source reasoning-focused artificial intelligence (AI) model, marking its first open-source initiative since the launch of GPT-2 in November 2019.
The company, headquartered in San Francisco, aims to unveil the model within the next few months, with a strong emphasis on safety throughout its development process.
Seeking to better understand the needs of developers, OpenAI is actively gathering feedback from the AI community. The move comes amid increasing competition within the open-source AI space, where companies like Meta, Mistral, Alibaba, Google, and Microsoft have already introduced multiple open models.
Despite its initial contributions with GPT-2, OpenAI has largely concentrated on releasing closed software that remains inaccessible for independent research or commercial modification. This has positioned the company as somewhat of an outlier in the rapidly evolving open-source landscape.
Earlier this year, during a Reddit AMA session, OpenAI CEO Sam Altman acknowledged the company’s shortcomings in its approach to open-source AI. He admitted that OpenAI has been “on the wrong side of history” and needs to rethink its strategy, though he noted that open-source efforts have not been the company's highest priority.
On Monday, Altman hinted at the company's shift in approach by sharing an “Open model feedback” form on X (formerly known as Twitter). The post signaled that OpenAI is now planning to build an advanced open-weight AI model designed for reasoning.
Unlike previous releases, OpenAI’s upcoming model will include open weights - numerical values representing connections made by the AI to generate outputs. However, without publicly accessible datasets and training techniques, replicating or building upon the model remains challenging. Such models are often considered only partially open-source.
Altman emphasised that safety will remain a key consideration, especially since the model may be modified after its release. OpenAI plans to use its Preparedness Framework to guide the development and monitoring of the model.