OpenAI is expected to release its secretive Project Strawberry AI reasoning model by the end of the month, making it available inside ChatGPT. This is earlier than initially reported, with subscribers of the $20 monthly ChatGPT Plus getting it first.
The update comes via The Information, with people close to the project telling the publication it will be available as a standalone option inside ChatGPT, likely via the model drop-down menu.
Project Strawberry, previously known as Q*, is OpenAI’s reasoning model. Unlike existing models, it will give ChatGPT the ability to ‘think’ before it responds. This means it will take considerably longer to respond to a request, but it will be more likely to be accurate.
What is unclear is whether Strawberry will require an additional subscription, have heavy rate limits imposed or only be made available to a small number of Plus subscribers.
What we know about Project Strawberry
OpenAI Japan CEO revealed that GPT-Next will be 100x more powerful than GPT-4 at KDDI Summit 2024The model will reportedly use a smaller version of OpenAI's mysterious Project StrawberryThe boost in performance comes from better architecture, not a ton of computing resources pic.twitter.com/DMlAIEPjiDSeptember 4, 2024
Project Strawberry seems to be a new model or system capable of improved reasoning, including going online to find information in preparation for solving a particularly complex problem. Reports suggest each query will take much longer but have fewer hallucinations.
Essentially, Strawberry ‘thinks’ before it responds to a message. There is evidence that OpenAI has been testing this in ChatGPT on some users, offering up two responses and asking users to pick the one they prefer. One is taking much longer, and the other is happening quickly.
The Information report suggests the initial version will be text only, without vision or multimodal capabilities such as an understanding of speech or video. But, when working with text, it should offer more accurate results from a single prompt than current models.
Its improved reasoning capabilities also means it should be better at math and coding problems and provide strategies and responses that take more than one step to explain or complete. This is an early move into the idea of AI agents that perform tasks autonomously.
When will Strawberry be available in ChatGPT
Reports suggest we will see Strawberry appear as an extra option inside ChatGPT by the end of this month. Some early suggestions were that it would never be public but rather a developer tool to help fine-tune and train other models, but it now seems we will get access.
It is unlikely to be available to those using the free version of ChatGPT due to the extra costs involved in performing multiple complex autonomous tasks before responding. There is also a suggestion that heavy rate limits will be even on the paid plan.
It also won’t be perfect. During testing, it sometimes struggled with remembering preferences and memory from previous ChatGPT conversations or took too long to answer even simple queries. At least that is according to people speaking to The Information.
Recently, OpenAI changed the model menu inside ChatGPT, relabelling GPT-4 as a "legacy model," whereas before, it was the best for complex tasks. This supports the idea we'll get a new option soon. It is unlikely we’ll get GPT-4.5 or GPT-5 this side of the election.
