August 8, 2012 - At a launch event held at 1 a.m. Beijing time today, theOpenAI The company officially announced GPT-5 Model.
OpenAI CEO Sam Altman was the first to take the stage, single-handedly announcing the launch of the GPT-5 model and stating that it's an important step towards generalized artificial intelligence (AGI). Talking to GPT-5, he argued, is like talking to a human expert.
The OpenAI team took the stage to introduce the GPT-5 model, stating that GPT-5 has set a new level on several benchmarks, including Swe Bench, and that the key areas of focus for GPT-5 are to improve reliability and accurately state facts.
OpenAI has announced that it will begin rolling out GPT-5 today for Free, Plus, and Pro subscribers, who will be able to use GPT-5 for free, subject to certain quota limitations, while Plus and Pro subscribers will be able to take advantage of higher usage benefits.
The first demo shown in the livestream asked the model to explain the Bernoulli effect, and the GPT-5 model responded quickly. A follow-up asked GPT-5 to create a visualization, and OpenAI explained that the response took longer.
The user can also ask GPT-5 to " think deeper" or " be more precise" in the actual prompt, and the user can see GPT-5's thought process as he or she answers the prompt.
GPT-5 performed well in OpenAI's programming demo, outputting over 200 lines of code in a few minutes, creating a website with many visual and even audio elements, designed to help a companion user learn French.
During this broadcast, the focus was on using the GPT-5 model for programming, but the team also indicated that GPT-5 will soon be available in voice mode, demonstrating ChatGPT's ability to respond with just one word in voice mode, summarizing pride and bias as "interpersonal relationships.
ChatGPT's Memory feature now supports Pro, Plus, and Teams subscribers to connect ChatGPT to their Google accounts to access calendar content. This means that ChatGPT can link to Gmail and Google Calendar, giving GPT-5 more contextual capabilities based on a user's daily schedule.
ChatGPT, which can even remind users to reply to their incoming emails, will be rolled out to Pro users first next week, with expansion to Plus and Teams subscribers to follow.
OpenAI also focuses on improving security and preventing spoofing. gPT-5 aims to help the user to the greatest extent possible within the allowed security limits. In contrast, in the case of o3, for example, where previous models would refuse to answer for security reasons, GPT-5 explains why help cannot be provided and directs the user to a third party or even a manual.
GPT-5 series has three models, GPT-5, GPT-5 mini and GPT-5 nano, the context length is 400K, the maximum output is 128K tokens, and the API price is as follows:
- GPT-5 $1.25 per million tokens input, $10 per million tokens output
- GPT-5 mini $0.25 per million tokens input, $2 per million tokens output
- GPT-5 nano $0.05 per million tokens input, $0.40 per million tokens output