Openi Creating a small, more efficient version of your clever artificial intelligence Free available models because it wants to respond to promotion and enthusiasm around the offering of a newly opened source from Chinese AI Startup Deepsek.
Wired told earlier that Openai was preparing new modelsThe O-3-Mine is called for release on 31 January. Researchers of the company have worked overtime to prepare it for prime time, according to sources talking about the situation of anonymity.
The O3-Min, which was teased Openai in December, is a small version of the model, which fulfills the most advanced AI argument capabilities of any Openai offering. The model component can break up difficult problems in parts to detect how to solve them.
The company said, “These powerful and fast models can achieve small models, leading its boundaries.” In a blog post Announcement of availability of o3-mini.
Openai O3-Min is providing all the plus, team and pro for Pro Categpt's Pro users. Users of the free version of Chatgpt will also be able to try O3-Mini, but will not be able to send the same questions, the company says.
Openai is clearly using PhD students to help train a new model for some time. Several weeks ago, the company began recruiting PhD computer science students at $ 100 per hour for a “research cooperation”, which would include “working on unpublished models” according to an email seen by Wired. “
Openai also appears that PhD students have been recruited with expertise in other areas through a company called Mercer that it uses to find employees for regular model training. recently job posting From Mercers on LinkedIn: “The overall goal of this project that can become a part of you, is that realistic scientific research is challenging to test the capabilities of large language models in generating codes to solve problems To create scientific coding questions. “
Job posting proceeds to give an example problem that is similar to a problem in a benchmark designed to test the ability of a large language model to solve complex science problems of a large language model called Scicocode. Has gone.
News comes Deepsek's R1 The US continues to cry the tech industry. The fact that such a powerful model can be released for free, pressures Google and anthropic to reduce their prices.
Openai is particularly eager to demonstrate that it is at the forefront of the development and commercialization of AI, according to sources inside the company.
The independent available models of Deepsak include those innovations that made it more efficient for both train and service. The company develops it using much less resources than OpenIAI and other American companies, currently the Frontier AI model, although the exact details of Deepsek's expenses are unknown. Openi Says it assumes R1 may include the output from its model in its training.
Found a tip?
Are you a current or former employee in Openai? We would like to hear from you. A non -work will be on the contact night using a phone or computer will_knight@wired.com Or on the signal via your user name wak01.
Openai's latest model cannot exclude R1 in terms of price, but it shows that the company will carry forward the efficiency share of its focus. Openai also states that the model is particularly strong in mathematics, science and coding.
The company says that the latest models will also include new features, including the ability to tap in web discovers, call function from the user's code, and to togue between various logic levels that speed for problem-co-ordination capabilities Do business from
The sudden growth of Deepsek has also raised questions about the US government's strategy to curb China's rise in AI. The last two US administration has introduced several restrictions to curb China's ability to reach China's most advanced NVidia chips, usually used for the manufacture of state -of -the -art AI models. Deepsek described many types of Nvidia chips in his research, but it is not clear what was exactly used.