Last Thursday, OpenAI unveiled the GPT-6 Astra model, which is touted on its official website as the "world's most intelligent and aligned model," without explicitly stating the attainment of AGI (Artificial General Intelligence). On the same day, NVIDIA's CEO, Jensen Huang, took to the X platform, proclaiming the advent of AGI and disclosing that Astra was trained utilizing approximately 100,000 NVIDIA GPUs, with plans to deploy an additional 400,000 GPUs in the future. The ARC Prize Foundation, which oversees the ARC-AGI-3 test, highlighted that while Astra achieved a 99.9% score in OpenAI's tailored testing environment, it only managed a 62.7% score in a standardized setting. The foundation clarified that attaining a perfect score on this benchmark does not necessarily signify the realization of AGI, although it did acknowledge that Astra surpasses human operational efficiency by a significant margin.

At a press conference, OpenAI President Greg Brockman remarked, "Welcome to the AGI era," but he underscored that this was his personal assessment, given the ambiguity surrounding the definition of AGI, which will ultimately be shaped by user consensus. Previously, OpenAI CEO Sam Altman had criticized the existing definition of AGI as inadequate, and the official website refrained from mentioning AGI. According to the agreement between OpenAI and Microsoft, the verification of AGI must be carried out by an independent panel of experts, and current public assertions about AGI do not constitute formal acknowledgment under the contract.

Due to the absence of a universally accepted definition of AGI, model companies, chip giants, benchmarking agencies, and investors continue to harbor differing opinions, and a widely recognized standard for AGI validation remains elusive.