OpenAI has informed Bloomberg that it has established an internal scale to measure the progression of its large language models towards artificial general intelligence (AGI), which refers to AI systems with human-like intelligence. Currently, chatbots like ChatGPT are categorized as Level 1, while OpenAI claims to be approaching Level 2, which involves a system capable of solving fundamental problems at the proficiency of a person holding a PhD. Moving forward, Level 3 denotes AI agents capable of taking actions on behalf of users, while Level 4 encompasses AI capable of generating new innovations. Finally, Level 5 represents the ultimate achievement of AGI, in which AI can perform the work of entire organizations. OpenAI has previously defined AGI as "a highly autonomous system surpassing humans in most economically valuable tasks. " OpenAI's mission of achieving AGI shapes its unique organizational structure, and their definition of AGI is of pivotal importance. The company has indicated that if another project with aligned values and safety consciousness comes close to building AGI before OpenAI does, they commit to not compete with that project and instead provide assistance. Although the wording in OpenAI's charter remains somewhat ambiguous, allowing leeway for decision-making by the for-profit entity overseen by the nonprofit, the development of a grading scale could help ascertain when AGI has been reached more definitively, thereby enabling OpenAI and its competitors to evaluate their progress. However, achieving AGI still seems a distant prospect, requiring substantial investment in computing power amounting to billions of dollars, if it is achievable at all. Various experts, including those at OpenAI, have offered widely varying timelines.

In October 2023, OpenAI CEO Sam Altman estimated that AGI is roughly "five years, give or take" away. The introduction of this new grading scale, which is still in the developmental stages, comes on the heels of OpenAI's collaboration announcement with Los Alamos National Laboratory. The partnership aims to explore the safe implementation of advanced AI models, such as GPT-4o, in bioscientific research. A program manager from Los Alamos, instrumental in securing the OpenAI partnership and overseeing the national security biology portfolio, articulated that the objective is to assess the capabilities of GPT-4o and establish a set of safety standards and other factors for the US government. In due course, both public and private AI models can be evaluated against these benchmarks. In May, OpenAI disbanded its safety team following the departure of its leader and co-founder, Ilya Sutskever. Shortly after, Jan Leike, a prominent OpenAI researcher, resigned and expressed concerns in a post that the company's focus on developing impressive products had overshadowed safety culture and processes. Although OpenAI refuted these claims, apprehensions remain regarding the company's approach should AGI become a reality. OpenAI has not disclosed the details of how it assigns models to the internal levels mentioned above and declined to comment on the matter when approached by The Verge. Nonetheless, during a recent all-hands meeting, company leaders demonstrated a research project employing the GPT-4 AI model, which Bloomberg reports exhibits new capabilities showcasing human-like reasoning.