In an anticipated move that has captured the attention of the tech community, OpenAI is preparing to launch an open-source artificial intelligence model next summer. This model is expected to be available for free download, marking the first time in approximately five years that the company has done so.
According to a report published by the tech website TechCrunch, the new model is anticipated to surpass other open models from companies like Meta and DeepSeek.
Despite the fact that full technical details have not yet been revealed, sources close to the project have indicated that OpenAI is considering adding an unprecedented feature to this model. Efforts are underway to enable it to communicate with stronger models hosted by the company in the cloud, with the aim of assisting it in processing complex inquiries.
This technology is known internally as "Task Offloading". The company's CEO, Sam Altman, has reportedly discussed it during a special meeting with developers from the open-source software community.
The idea for this feature first emerged from a suggestion made by one of the developers during a listening event organized by OpenAI. Since that time, it has begun to receive increasing internal attention.
If implemented, this step could potentially open the door for the new model to call the company's APIs when higher computational power is needed. However, it is not yet known whether it will be allowed to utilize tools such as internet search or image generation.
The proposed mechanism bears resemblance to what Apple offers within its "Apple Intelligence" system. That system relies on a combination of local models and those operating in private data centers.
On the other hand, the new feature might contribute to attracting the open-source software community towards OpenAI's paid environment.
Despite these ambitions, numerous questions still remain. These questions are especially pertinent regarding the pricing mechanism, usage limits, and the extent of reliance on cloud infrastructure.
Based on the available information, OpenAI is training the model from scratch. This means it is not relying on any previous version.
Expectations suggest that its performance will be lower than the company's currently strongest model, "o3". Nevertheless, it might outperform DeepSeek's "R1" in certain benchmark tests.
