OpenAI, the global leader in artificial intelligence, appears poised for a significant development after a period of relative quiet. According to a report from The Information, citing multiple sources, the company plans to officially launch a new AI model, codenamed “Strawberry,” in about two weeks. This model will focus on enhanced reasoning capabilities and will be integrated into the ChatGPT service. With enthusiastic promotion from OpenAI’s CEO Sam Altman and a lack of exciting news from the company lately, anticipation is building around this new offering.
The “Strawberry” model stands out for its emphasis on reasoning. It is designed to handle complex, multi-step inquiries more effectively and intuitively than the current GPT-4o model. Users often find themselves needing to provide detailed guidance on every reasoning step when using GPT-4o for multi-step problems. In contrast, “Strawberry” aims to simplify this process, improving users’ experiences and outcomes, even potentially solving New York Times crossword puzzles.
In addition to its strengths in mathematics and programming, “Strawberry” is also capable of tackling broader and more subjective questions. For instance, if prompted to brainstorm a marketing strategy, “Strawberry” is expected to provide responses that are contextually relevant and detail-oriented, such as generating a week-by-week execution plan.
The core difference in response quality comes from “Strawberry” taking time to “think” before answering. It may ask follow-up questions, leading to more comprehensive responses. However, this feature could be a potential drawback. Reports suggest that the model may take between 10 to 20 seconds to begin delivering an answer, raising concerns among users in sectors where real-time interaction with AI is increasingly important.
Moreover, there are challenges concerning the model’s performance in certain scenarios. In cases where either GPT-4o or “Strawberry” could provide reasoning, users might find that “Strawberry” does not significantly outperform its predecessor, despite the longer wait time. Feedback indicates that users may prefer the quicker, more immediate responses from GPT-4o.
Additionally, during internal testing, “Strawberry” has faced issues with retaining conversation history. The model is currently limited to text input and output and has yet to overcome the crucial hurdle of multimodal capabilities, which would allow it to process images as well.
Beyond the model’s capabilities, there is also speculation regarding OpenAI’s commercial strategy. Earlier reports suggested that OpenAI is considering setting a higher price for the forthcoming “Strawberry” model, or possibly for its next flagship model, Orion. A more realistic strategy might involve introducing a new tier within ChatGPT Plus that offers unlimited access to “Strawberry” or increased response speeds. However, the success of this pricing strategy hinges on widespread user acceptance of the model’s performance enhancements.
Currently, the ChatGPT Plus subscription costs $20 per month for individual users, while enterprise users pay between $25 to $60 per account, depending on their needs. Recently, it was reported that OpenAI generates approximately $283 million per month from enterprise services and ChatGPT subscriptions, although the company continues to operate at a net loss.