OpenAI is expected to release its secretive Project Strawberry AI inference model by the end of the month and make it available within ChatGPT This is earlier than originally reported, and ChatGPT Plus subscribers, who pay $20/month, will get it first

The update comes via The Information, and people close to the project tell us that it will be available as a standalone option within ChatGPT, possibly from the model drop-down menu

Project Strawberry, formerly known as Q*, is an OpenAI inference model Unlike existing models, it gives ChatGPT the ability to “think” before responding This means that it takes considerably longer to respond to a request, but is more likely to be accurate

What is unclear is whether Strawberry will require additional subscriptions, impose heavy rate limits, or be offered only to a small number of Plus subscribers

Project Strawberry appears to be a new model or system that could improve reasoning, especially in preparation for solving complex problems, such as finding information online Reportedly, each query takes considerably longer, but there are fewer illusions

Basically, Strawberry “thinks” before responding to a message; there is evidence that OpenAI has conducted this test on ChatGPT for several users One took quite a while, the other happened immediately

Information reports that the initial version is text-only, with no visual or multimodal features such as speech or video understanding However, when dealing with text, it should produce more accurate results from a single prompt than the current model

Also, improved reasoning capabilities would mean that AI would be better at math and coding problems, providing strategies and answers that take more than one step to explain or complete This is an early move toward the idea of AI agents that perform tasks autonomously

Reports indicate that Strawberry will be an additional option for ChatGPT by the end of this month Early proposals said it would not be publicly available, but rather a developer tool to help tweak and train other models, but now it appears we will have access to it

It is unlikely to be available to those using the free version of ChatGPT due to the extra cost of performing multiple complex autonomous tasks before responding It has also been suggested that even paid plans have heavy rate limits

Also, it is not perfect During testing, we had trouble recalling preferences and memories from previous ChatGPT conversations, and sometimes it took too long to answer simple inquiries At least, this was the story of those interviewed by The Information

Recently, OpenAI changed the model menu within ChatGPT, renaming GPT-4, previously best suited for complex tasks, the “legacy model” This supports the idea that we will soon have new options; GPT-45 or GPT-5 will not be available during the election