OpenAI Launches "Smartest' Model in Company's History

OpenAI, a pioneer and leader in generative artificial intelligence since the launch of ChatGPT in late 2022, last week unveiled GPT-5, described as its "smartest,” "fastest,” and "most useful” model. The launch came amid fierce competition among tech giants to develop general artificial intelligence with cognitive abilities superior to those of humans.

• From "high school student” to "doctoral expert”

"GPT-5 is like talking to a doctoral-level expert on any subject,” said Sam Altman, CEO and co-founder of OpenAI, during a press conference.

Altman made an evolutionary comparison: GPT-3, used in the first version of ChatGPT, resembled "a high school student who sometimes gives the right answer, and sometimes says anything,” while GPT-4 was comparable to "a college student.” The new model, he says, makes a significant qualitative leap in terms of reasoning and autonomy in completing tasks.

• The Global Race for Generative AI

Big companies-Google (Gemini), Meta (Meta AI), Anthropic (Claude), and others-are constantly releasing more and more advanced models. Their goal is clear: to transform virtual assistants into indispensable tools for users' daily lives and for developers. Currently, ChatGPT remains the most well-known name in the field, with nearly 700 million weekly active users. OpenAI aims to maintain its supremacy through continuous innovation.

• New features and increased customization

GPT-5 comes with a number of user-oriented features. The assistant can be customized to adopt a concise, friendly or even sarcastic tone and will be able to connect directly to services such as Gmail. "GPT-5 can instantly create software on demand (...) it is an incredible superpower,” said Altman. An example presented at the conference: engineer Yann Dubois asked the model, in natural language, to develop an online application for learning French through games. In a few minutes, GPT-5 generated hundreds of lines of code and a functional, albeit simple, website.

According to OpenAI, GPT-5 "hallucinates” - that is, gives false answers that appear to be true - much less often than previous versions and can recognize when it does not know an answer. "We trained it to be honest (...) so that users do not lie,” stressed Alex Beutel, responsible for the company's product safety.

In terms of security, the approach has become more nuanced: if there are suspicions of criminal use, the model only provides general information, which cannot be used to cause harm.

• Still a long way from general artificial intelligence

While GPT-5's performance is impressive, Altman said the model has not yet reached the status of general artificial intelligence. "It's not a model that continuously learns from what it discovers,” he explained.

OpenAI, founded in 2015 as a non-profit organization, has been committed to developing general AI for the benefit of all humanity since its inception. The launch of GPT-5 is a major, but not final, step in that direction.