OpenAI has recently shared that retraining its powerful model GPT-4 now requires just a small team of five to ten people. This advancement is due to the breakthroughs achieved with its latest model, GPT-4.5. Sam Altman, CEO of OpenAI, shared this exciting development during a company podcast.
From Hundreds of People to a Small Team
In the past, building GPT-4 was a massive task involving hundreds of engineers, making it a highly resource-intensive project. However, OpenAI now claims that retraining GPT-4 would require a much smaller team, thanks to the knowledge and innovations learned from the creation of GPT-4.5.
GPT-4.5: A Game-Changing Model
Released in February 2025, GPT-4.5 is OpenAI’s most advanced model to date, marking a significant milestone in AI research. It is described as being “10x smarter” than GPT-4, with improvements in data efficiency and algorithm performance. Alex Paino, who led pre-training for GPT-4.5, explained that retraining GPT-4 was made simpler by leveraging the insights gained from the GPT-4.5 research program.
The New Era of AI: No More “Compute Constraints”
Sam Altman also emphasized that OpenAI is no longer limited by computing power. For years, the availability of computational resources was one of the biggest challenges in AI development. However, with ongoing advancements and massive investments from companies like Microsoft, Amazon, Google, and Meta, OpenAI is now free from these constraints.
The Shift in AI Infrastructure
The push for AI progress has seen tech giants investing billions into AI infrastructure. In March 2025, OpenAI announced its largest-ever private funding round, including $30 billion from SoftBank and $10 billion from other investors. This investment will help OpenAI continue to expand its computing power and accelerate the development of more advanced AI models.
Data Efficiency: The Next Frontier
While compute power has been a major hurdle, OpenAI researchers believe that data efficiency is the next challenge to overcome. Daniel Selsam, a researcher at OpenAI, explained that the key to future advancements lies in better utilizing available data. As compute power increases, the growth of data slows down, creating a bottleneck. To push past this limitation, algorithmic innovations are necessary to maximize the value of existing data.
The Road Ahead for AI Models
The progress made with GPT-4.5 has opened new possibilities for AI development, but there are still challenges to tackle. The next steps will involve refining algorithms and making better use of data to ensure that AI can scale even further. OpenAI’s breakthroughs have shown the world that AI’s potential is far from being fully realized, and with continued innovation, we can expect even more impressive developments in the near future.
