news

key points:

tencent technology news according to foreign media reports, people familiar with the matter revealed on tuesday that as part of the chatgpt service, openai is about to launch a new reasoning artificial intelligence model called "strawberry". the launch time of "strawberry" may be earlier than the original autumn, and is expected to meet the public in the next two weeks.

although the final release date of strawberry is still subject to change, two people who have experienced the model have revealed some information about the new features to foreign media. although this inference ai model is part of chatgpt, it is an independent product. the specific delivery method has not yet been clarified, but it may appear in the form of an ai model selection menu in chatgpt. compared with the standard service, the strawberry model is better in some aspects, but it also has some limitations.

the most striking feature of strawberry is that it "thinks" before answering questions, rather than answering immediately, a process that usually lasts 10 to 20 seconds. this thinking ability is the main difference between it and other conversational ai models. however, the first version of strawberry can only process text and does not support images, which means that it has not yet reached the multimodal level of most current large language models. this may beOpenAIthe result of rushing to launch a product under competitive pressure.

in terms of pricing, strawberry may be different from the free and subscription models of openai chatbots. although the specific pricing strategy has not yet been determined, people familiar with the matter revealed that strawberry may implement rate limits, that is, limit the number of messages sent by users per hour, and may provide faster response services at higher prices. this strategy may attract more users to pay for the new model.

it is expected that paid chatgpt users will be the first to experience this new model before strawberry is open to the general free users. it is unclear whether openai will charge higher fees for users of the more advanced version of strawberry.

strawberry is expected to be more efficient than gpt-4o in processing complex or multi-step queries. when using chatgpt, users often need to enter additional instructions, such as asking the chatbot to show its reasoning process in order to get the final answer. strawberry is designed to reduce this need and help users get intelligent results more easily.

strawberry not only performs better in solving math problems and programming tasks, but also better handles more subjective business tasks, such as developing product marketing strategies. in such tasks, strawberry is able to provide more specific and detailed suggestions, such as developing a detailed weekly execution plan. in addition, strawberry's thinking stage helps reduce errors and is more likely to ask users follow-up questions when necessary to ensure that the user's questions can be fully answered.

however, openai may need to iron out some details before or after release. for example, although strawberry is theoretically able to skip thinking steps when faced with simple questions, it doesn’t always do so in practice. this could cause it to spend too much time answering questions that other openai models can handle quickly.

some users who have experienced the prototype of strawberry complained that although the answers of strawberry are slightly better than the gpt-4o currently provided by openai, it is not worth waiting for 10 to 20 seconds. in addition, although strawberry is designed to remember and combine previous conversations with customers to answer questions, especially when users have specific preferences (such as wanting software code to be written in a specific format), the prototype sometimes has difficulties in this regard.

although openai is at the forefront of products driven by large language models, it faces increasing competition. for example, google beat openai to the punch last month by launching an ai-powered voice assistant that is able to flexibly respond to user interruptions and sudden changes in the topic. openai demonstrated the voice assistant gpt-4o voice in may this year, but later postponed the release time in order to strengthen security measures, such as ensuring that inappropriate content can be rejected. if "strawberry" is released smoothly, it will hopefully help openai regain the strong momentum it has maintained for most of the past two years. (compiled by wuji)