OpenAI is constantly working on developing new models, with several teams exploring different approaches to achieve its long-term goal of artificial intelligence (AGI).

According to Reuters, there is a new project codenamed “Strawberry,” which some X insiders predict may be a new version of the infamous “inference mode” Q* revealed by a leak last year.

Project Strawberry appears to be a new model or system that could improve reasoning, such as looking for information online to solve particularly complex problems.

What is not clear is whether Strawberry is the model that OpenAI CTO Mira Murati was talking about when she said that the next generation of AI will be as intelligent as a person with a PhD, or whether this is OpenAI reaching “Level 2” on the path to AGI, or reasoners. The question is whether this is evidence that we have reached “level 2” on the path to AGI, i.e., the reasoner.

The Reuters report, based on internal documents and comments from insiders, suggests that this could be a significant upgrade in artificial intelligence capabilities.

According to that official, Strawberry is even a “closely guarded secret” within OpenAI, but while it may become part of ChatGPT in the future, its main purpose is apparently to conduct “deeper research” beyond simple user queries.

Some of the features include being able to plan well enough when responding to navigate the Internet autonomously and reliably without being directed by the user.

ChatGPT can already search the web for specific details, but not reliably, and often needs to be specifically instructed to do so to look for live data as well as to look at study data.

Open AI is not directly mentioned in the reports about Strawberry, but according to Reuters, it is common for the company to conduct ongoing research on various new AI features.

Project Strawberry is only an Open AI research project, and it is not certain whether it will see the light of day. But even if it does not become a product per se, the research will likely feed into future AI models trained by OpenAI and updates to future products.

Improved inference will be incorporated into GPT-5 training, and Project Strawberry may become a new style of model that suddenly appears and changes the game, like Omni; unlike GPT-4, GPT-4o (Omni) was native and multimodal, while GPT-5o (Omni) was not.

Strawberry could be a new type of model, a native reasoner from the ground up, requiring no tweaking or additional training. Little is actually known about it, but OpenAI's comments to Reuters suggest that it is an important requirement for AI to understand it like a human.