OpenAI is reportedly working on launching its new AI model “Strawberry,” which has better mathematical performance than GPT-4o.

OpenAI is reportedly preparing to launch a new AI model, known internally as “Strawberry.” The model would be able to solve math problems it has never encountered before. In addition, it can perform sophisticated tasks and even solve word puzzles, The Information reports. According to previous sources, the model performs better than GPT-4o in the MATH benchmarks. The launch is expected to take place in the fall.

Strawberry model

The new AI model was previously known internally as Q* but has now been renamed Strawberry. The model would be introduced this fall as part of a chatbot, likely ChatGPT, two people involved told The Information.

Compared to traditional language models, Strawberry can solve new problems and tasks that are beyond the reach of current AI models. For example, the new model was able to solve the New York Times’ “Connections” word puzzle.

According to previous claims, the Strawberry model scores more than 90 percent on the MATH benchmark, which is a collection of mastery-level mathematical problems. Currently, GPT-4o leads with the highest MATH benchmark score (76.6%). If the Strawberry model delivers on its promises, it will dethrone GPT-4o.