OpenAI plans to release “Strawberry,” its new reasoning-focused AI model, as part of the ChatGPT service in the coming weeks, according to a report by The Information. This new model is expected to offer a distinct experience compared to OpenAI’s existing lineup, with a particular emphasis on handling complex tasks that require deeper reasoning.
Strawberry’s standout feature is its ability to “think” before delivering responses, taking between 10 to 20 seconds to process a query before answering. This deliberate pause is designed to automate a sort of chain-of-thought prompting that should improve the accuracy of responses. This is especially valuable for more complex or multi-step queries, where users might otherwise need to guide the model through detailed instructions or use various hacks. This kind of reasoning capability is expected to streamline tasks like coding and solving intricate math problems, but also lend itself to business tasks, such as generating detailed marketing strategies or week-by-week execution plans.
Although Strawberry will be integrated into ChatGPT, it will function as a separate offering. Users may be able to select it from a dropdown menu within the service, though specific details around how it will be offered remain unclear. Importantly, this version of Strawberry will only handle text-based queries.
While pricing details haven’t been confirmed, there are suggestions that OpenAI could implement rate limits, restricting the number of messages users can send per hour, with higher-priced tiers offering faster response times. Early access is anticipated to go to paying ChatGPT customers before the model becomes available to free-tier users.
Despite its promise however, early testers have reported mixed results. Some users felt that the 10 to 20 seconds of extra processing didn’t always yield significantly better results compared to existing options like GPT-4o. Others noted inconsistencies in the model's ability to retain context from previous interactions or following custom instructions.
Most importantly, there is a big question around how many users will actually get access to Strawberry upon release. OpenAI has a track record of announcing features like multimodal video support, the GPT-4o voice engine, and the Sora video generator, which remain accessible to only a small subset of users long after their initial unveiling.
Either way, the coming weeks will reveal whether Strawberry is all it has been hyped up to be–and crucially, whether it will be generally available to many users.